def load_yolov2(image,output_file,dataset='coco',threshold=0.5,nms_thresh=0.4):
batch_size = 1
confidence = threshold
start = 0
global num_classes
imlist = image
output_file_names = [output_file]
CUDA = torch.cuda.is_available()
if dataset == "pascal":
inp_dim = 416
num_classes = 20
classes = load_classes('data/voc.names')
weightsfile = 'yolov2-voc.weights'
cfgfile = "cfg/yolo-voc.cfg"
elif dataset == "coco":
inp_dim = 544
num_classes = 80
classes = load_classes('data/coco.names')
weightsfile = 'yolov2.weights'
cfgfile = "cfg/yolo.cfg"
else:
print("Invalid dataset")
exit()
stride = 32
#Set up the neural network
print("Loading network.....")
model = Darknet(cfgfile)
model.load_weights(weightsfile)
print("Network successfully loaded")
#If there's a GPU availible, put the model on GPU
if CUDA:
model.cuda()
model(get_test_input(inp_dim, CUDA))
#Set the model in evaluation mode
model.eval()
read_dir = time.time()
#Detection phase
load_batch = time.time()
batches = list(map(prep_image, imlist, [inp_dim for x in range(len(imlist))]))
im_batches = [x[0] for x in batches]
orig_ims = [x[1] for x in batches]
im_dim_list = [x[2] for x in batches]
im_dim_list = torch.FloatTensor(im_dim_list).repeat(1,2)
if CUDA:
im_dim_list = im_dim_list.cuda()
leftover = 0
if (len(im_dim_list) % batch_size):
leftover = 1
if batch_size != 1:
num_batches = len(imlist) // batch_size + leftover
im_batches = [torch.cat((im_batches[i*batch_size : min((i + 1)*batch_size,
len(im_batches))])) for i in range(num_batches)]
i = 0
output = torch.FloatTensor(1, 8)
write = False
# model(get_test_input(inp_dim, CUDA))
start_det_loop = time.time()
for batch in im_batches:
#load the image
start = time.time()
if CUDA:
batch = batch.cuda()
prediction = model(Variable(batch, volatile = True))
prediction = prediction.data
#Apply offsets to the result predictions
#Tranform the predictions as described in the YOLO paper
#flatten the prediction vector
# B x (bbox cord x no. of anchors) x grid_w x grid_h --> B x bbox x (all the boxes)
# Put every proposed box as a row.
#get the boxes with object confidence > threshold
#Convert the cordinates to absolute coordinates
prediction = predict_transform(prediction, inp_dim, stride, model.anchors, num_classes, confidence, CUDA)
if type(prediction) == int:
i += 1
continue
#perform NMS on these boxes, and save the results
#I could have done NMS and saving seperately to have a better abstraction
#But both these operations require looping, hence
#clubbing these ops in one loop instead of two.
#loops are slower than vectorised operations.
prediction = write_results(prediction, num_classes, nms = True, nms_conf = nms_thresh)
end = time.time()
# print(end - start)
prediction[:,0] += i*batch_size
if not write:
output = prediction
write = 1
else:
output = torch.cat((output,prediction))
for image in imlist[i*batch_size: min((i + 1)*batch_size, len(imlist))]:
im_id = imlist.index(image)
objs = [classes[int(x[-1])] for x in output if int(x[0]) == im_id]
print("{0:20s} predicted in {1:6.3f} seconds".format(image.split("/")[-1], (end - start)/batch_size))
print("{0:20s} {1:s}".format("Objects Detected:", " ".join(objs)))
print("----------------------------------------------------------")
i += 1
if CUDA:
torch.cuda.synchronize()
output_recast = time.time()
output[:,1:5] = torch.clamp(output[:,1:5], 0.0, float(inp_dim))
im_dim_list = torch.index_select(im_dim_list, 0, output[:,0].long())/inp_dim
output[:,1:5] *= im_dim_list
class_load = time.time()
colors = pkl.load(open("pallete", "rb"))
draw = time.time()
def write(x, batches, results):
c1 = tuple(x[1:3].int())
c2 = tuple(x[3:5].int())
img = results[int(x[0])]
cls = int(x[-1])
label = "{0}".format(classes[cls])
color = random.choice(colors)
cv2.rectangle(img, c1, c2,color, 1)
t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 1 , 1)[0]
c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4
cv2.rectangle(img, c1, c2,color, -1)
cv2.putText(img, label, (c1[0], c1[1] + t_size[1] + 4), cv2.FONT_HERSHEY_PLAIN, 1, [225,255,255], 1);
return img
list(map(lambda x: write(x, im_batches, orig_ims), output))
#det_names = pd.Series(imlist).apply(lambda x: "{}/det_{}".format(args.det,x.split("/")[-1]))
det_names = output_file_names
list(map(cv2.imwrite, det_names, orig_ims))
end = time.time()
print()
print("SUMMARY")
print("----------------------------------------------------------")
print("{:25s}: {}".format("Task", "Time Taken (in seconds)"))
print()
print("{:25s}: {:2.3f}".format("Reading addresses", load_batch - read_dir))
print("{:25s}: {:2.3f}".format("Loading batch", start_det_loop - load_batch))
print("{:25s}: {:2.3f}".format("Detection (" + str(len(imlist)) + " images)", output_recast - start_det_loop))
print("{:25s}: {:2.3f}".format("Output Processing", class_load - output_recast))
print("{:25s}: {:2.3f}".format("Drawing Boxes", end - draw))
print("{:25s}: {:2.3f}".format("Average time_per_img", (end - load_batch)/len(imlist)))
print("----------------------------------------------------------")
torch.cuda.empty_cache()
# Get the dataloader for test data
test_loader = torch.utils.data.DataLoader(dataset.listDataset(
testlist,
shape=(test_width, test_height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]),
train=False),
batch_size=1,
shuffle=False,
**kwargs)
# Pass the model to GPU
if use_cuda:
model = model.cuda(
) # model = torch.nn.DataParallel(model, device_ids=[0]).cuda() # Multiple GPU parallelism
# Get the optimizer
params_dict = dict(model.named_parameters())
params = []
for key, value in params_dict.items():
if key.find('.bn') >= 0 or key.find('.bias') >= 0:
params += [{'params': [value], 'weight_decay': 0.0}]
else:
params += [{'params': [value], 'weight_decay': decay * batch_size}]
optimizer = optim.SGD(model.parameters(),
lr=learning_rate / batch_size,
momentum=momentum,
dampening=0,
weight_decay=decay * batch_size)
def valid(datacfg, cfgfile, weightfile, outfile):
options = read_data_cfg(datacfg)
valid_images = options['valid']
name_list = options['names']
prefix = 'results'
names = load_class_names(name_list)
with open(valid_images) as fp:
tmp_files = fp.readlines()
valid_files = [item.rstrip() for item in tmp_files]
m = Darknet(cfgfile)
m.print_network()
m.load_weights(weightfile)
m.cuda()
m.eval()
valid_dataset = dataset.listDataset(valid_images,
shape=(m.width, m.height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]))
valid_batchsize = 2
assert (valid_batchsize > 1)
kwargs = {'num_workers': 4, 'pin_memory': True}
valid_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=valid_batchsize,
shuffle=False,
**kwargs)
fps = [0] * m.num_classes
if not os.path.exists('results'):
os.mkdir('results')
for i in range(m.num_classes):
buf = '%s/%s%s.txt' % (prefix, outfile, names[i])
fps[i] = open(buf, 'w')
lineId = -1
conf_thresh = 0.005
nms_thresh = 0.45
if m.net_name() == 'region': # region_layer
shape = (0, 0)
else:
shape = (m.width, m.height)
for _, (data, target, org_w, org_h) in enumerate(valid_loader):
data = data.cuda()
output = m(data)
batch_boxes = get_all_boxes(output,
shape,
conf_thresh,
m.num_classes,
only_objectness=0,
validation=True)
for i in range(len(batch_boxes)):
lineId += 1
fileId = os.path.basename(valid_files[lineId]).split('.')[0]
#width, height = get_image_size(valid_files[lineId])
width, height = float(org_w[i]), float(org_h[i])
print(valid_files[lineId])
boxes = batch_boxes[i]
correct_yolo_boxes(boxes, width, height, m.width, m.height)
boxes = nms(boxes, nms_thresh)
for box in boxes:
x1 = (box[0] - box[2] / 2.0) * width
y1 = (box[1] - box[3] / 2.0) * height
x2 = (box[0] + box[2] / 2.0) * width
y2 = (box[1] + box[3] / 2.0) * height
det_conf = box[4]
for j in range((len(box) - 5) // 2):
cls_conf = box[5 + 2 * j]
cls_id = box[6 + 2 * j]
prob = det_conf * cls_conf
fps[cls_id].write('%s %f %f %f %f %f\n' %
(fileId, prob, x1, y1, x2, y2))
for i in range(m.num_classes):
fps[i].close()
def valid(datacfg, cfgfile, weightfile, outfile):
def truths_length(truths):
for i in range(50):
if truths[i][1] == 0:
return i
# Parse configuration files
options = read_data_cfg(datacfg)
valid_images = options['valid']
meshname = options['mesh']
backupdir = options['backup']
name = options['name']
if not os.path.exists(backupdir):
makedirs(backupdir)
# Parameters
prefix = 'results'
seed = int(time.time())
gpus = '0' # Specify which gpus to use
test_width = 544
test_height = 544
torch.manual_seed(seed)
use_cuda = True
if use_cuda:
os.environ['CUDA_VISIBLE_DEVICES'] = gpus
torch.cuda.manual_seed(seed)
save = False
testtime = True
use_cuda = True
num_classes = 1
testing_samples = 0.0
eps = 1e-5
notpredicted = 0
conf_thresh = 0.1
nms_thresh = 0.4
match_thresh = 0.5
if save:
makedirs(backupdir + '/test')
makedirs(backupdir + '/test/gt')
makedirs(backupdir + '/test/pr')
# To save
testing_error_trans = 0.0
testing_error_angle = 0.0
testing_error_pixel = 0.0
errs_2d = []
errs_3d = []
errs_trans = []
errs_angle = []
errs_corner2D = []
preds_trans = []
preds_rot = []
preds_corners2D = []
gts_trans = []
gts_rot = []
gts_corners2D = []
edges_corners = [[0, 1], [0, 2], [0, 4], [1, 3], [1, 5], [2, 3], [2, 6],
[3, 7], [4, 5], [4, 6], [5, 7], [6, 7]]
# Read object model information, get 3D bounding box corners
mesh = MeshPly(meshname)
vertices = np.c_[np.array(mesh.vertices),
np.ones((len(mesh.vertices), 1))].transpose()
corners3D = get_3D_corners(vertices)
# diam = calc_pts_diameter(np.array(mesh.vertices))
diam = float(options['diam'])
# Read intrinsic camera parameters
internal_calibration = get_camera_intrinsic()
# Get validation file names
with open(valid_images) as fp:
tmp_files = fp.readlines()
valid_files = [item.rstrip() for item in tmp_files]
# Specicy model, load pretrained weights, pass to GPU and set the module in evaluation mode
model = Darknet(cfgfile)
model.print_network()
model.load_weights(weightfile)
model.cuda()
model.eval()
# Get the parser for the test dataset
valid_dataset = dataset.listDataset(valid_images,
shape=(test_width, test_height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]))
valid_batchsize = 1
# Specify the number of workers for multiple processing, get the dataloader for the test dataset
kwargs = {'num_workers': 4, 'pin_memory': True}
test_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=valid_batchsize,
shuffle=False,
**kwargs)
logging(" Testing {}...".format(name))
logging(" Number of test samples: %d" % len(test_loader.dataset))
# Iterate through test batches (Batch size for test data is 1)
count = 0
z = np.zeros((3, 1))
for batch_idx, (data, target) in enumerate(test_loader):
# Images
img = data[0, :, :, :]
img = img.numpy().squeeze()
img = np.transpose(img, (1, 2, 0))
t1 = time.time()
# Pass data to GPU
if use_cuda:
data = data.cuda()
target = target.cuda()
# Wrap tensors in Variable class, set volatile=True for inference mode and to use minimal memory during inference
data = Variable(data, volatile=True)
t2 = time.time()
# Forward pass
output = model(data).data
t3 = time.time()
# Using confidence threshold, eliminate low-confidence predictions
all_boxes = get_region_boxes(output, conf_thresh, num_classes)
t4 = time.time()
# Iterate through all images in the batch
for i in range(output.size(0)):
# For each image, get all the predictions
boxes = all_boxes[i]
# For each image, get all the targets (for multiple object pose estimation, there might be more than 1 target per image)
truths = target[i].view(-1, 21)
# Get how many object are present in the scene
num_gts = truths_length(truths)
# Iterate through each ground-truth object
for k in range(num_gts):
box_gt = [
truths[k][1], truths[k][2], truths[k][3], truths[k][4],
truths[k][5], truths[k][6], truths[k][7], truths[k][8],
truths[k][9], truths[k][10], truths[k][11], truths[k][12],
truths[k][13], truths[k][14], truths[k][15], truths[k][16],
truths[k][17], truths[k][18], 1.0, 1.0, truths[k][0]
]
best_conf_est = -1
# If the prediction has the highest confidence, choose it as our prediction for single object pose estimation
for j in range(len(boxes)):
if (boxes[j][18] > best_conf_est):
match = corner_confidence9(
box_gt[:18], torch.FloatTensor(boxes[j][:18]))
box_pr = boxes[j]
best_conf_est = boxes[j][18]
# Denormalize the corner predictions
corners2D_gt = np.array(np.reshape(box_gt[:18], [9, 2]),
dtype='float32')
corners2D_pr = np.array(np.reshape(box_pr[:18], [9, 2]),
dtype='float32')
corners2D_gt[:, 0] = corners2D_gt[:, 0] * 640
corners2D_gt[:, 1] = corners2D_gt[:, 1] * 480
corners2D_pr[:, 0] = corners2D_pr[:, 0] * 640
corners2D_pr[:, 1] = corners2D_pr[:, 1] * 480
preds_corners2D.append(corners2D_pr)
gts_corners2D.append(corners2D_gt)
# Compute corner prediction error
corner_norm = np.linalg.norm(corners2D_gt - corners2D_pr,
axis=1)
corner_dist = np.mean(corner_norm)
errs_corner2D.append(corner_dist)
# Compute [R|t] by pnp
R_gt, t_gt = pnp(
np.array(np.transpose(
np.concatenate((np.zeros((3, 1)), corners3D[:3, :]),
axis=1)),
dtype='float32'), corners2D_gt,
np.array(internal_calibration, dtype='float32'))
R_pr, t_pr = pnp(
np.array(np.transpose(
np.concatenate((np.zeros((3, 1)), corners3D[:3, :]),
axis=1)),
dtype='float32'), corners2D_pr,
np.array(internal_calibration, dtype='float32'))
if save:
preds_trans.append(t_pr)
gts_trans.append(t_gt)
preds_rot.append(R_pr)
gts_rot.append(R_gt)
np.savetxt(
backupdir + '/test/gt/R_' + valid_files[count][-8:-3] +
'txt', np.array(R_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/gt/t_' + valid_files[count][-8:-3] +
'txt', np.array(t_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/R_' + valid_files[count][-8:-3] +
'txt', np.array(R_pr, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/t_' + valid_files[count][-8:-3] +
'txt', np.array(t_pr, dtype='float32'))
np.savetxt(
backupdir + '/test/gt/corners_' +
valid_files[count][-8:-3] + 'txt',
np.array(corners2D_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/corners_' +
valid_files[count][-8:-3] + 'txt',
np.array(corners2D_pr, dtype='float32'))
# Compute translation error
trans_dist = np.sqrt(np.sum(np.square(t_gt - t_pr)))
errs_trans.append(trans_dist)
# Compute angle error
angle_dist = calcAngularDistance(R_gt, R_pr)
errs_angle.append(angle_dist)
# Compute pixel error
Rt_gt = np.concatenate((R_gt, t_gt), axis=1)
Rt_pr = np.concatenate((R_pr, t_pr), axis=1)
proj_2d_gt = compute_projection(vertices, Rt_gt,
internal_calibration)
proj_2d_pred = compute_projection(vertices, Rt_pr,
internal_calibration)
proj_corners_gt = np.transpose(
compute_projection(corners3D, Rt_gt, internal_calibration))
proj_corners_pr = np.transpose(
compute_projection(corners3D, Rt_pr, internal_calibration))
norm = np.linalg.norm(proj_2d_gt - proj_2d_pred, axis=0)
pixel_dist = np.mean(norm)
errs_2d.append(pixel_dist)
# Visualize
fig = plt.figure()
plt.xlim((0, 640))
plt.ylim((0, 480))
plt.imshow(scipy.misc.imresize(img, (480, 640)))
# Projections
for edge in edges_corners:
plt.plot(proj_corners_gt[edge, 0],
proj_corners_gt[edge, 1],
color='g',
linewidth=3.0)
plt.plot(proj_corners_pr[edge, 0],
proj_corners_pr[edge, 1],
color='b',
linewidth=3.0)
plt.gca().invert_yaxis()
# plt.show()
plt.savefig(outfile + '/output_' + str(count) + '_.png',
bbox_inches='tight')
fig.canvas.draw()
count = count + 1
# Compute 3D distances
transform_3d_gt = compute_transformation(vertices, Rt_gt)
transform_3d_pred = compute_transformation(vertices, Rt_pr)
norm3d = np.linalg.norm(transform_3d_gt - transform_3d_pred,
axis=0)
vertex_dist = np.mean(norm3d)
errs_3d.append(vertex_dist)
# Sum errors
testing_error_trans += trans_dist
testing_error_angle += angle_dist
testing_error_pixel += pixel_dist
testing_samples += 1
count = count + 1
t5 = time.time()
# Compute 2D projection error, 6D pose error, 5cm5degree error
px_threshold = 5
acc = len(np.where(
np.array(errs_2d) <= px_threshold)[0]) * 100. / (len(errs_2d) + eps)
acc5cm5deg = len(
np.where((np.array(errs_trans) <= 0.05)
& (np.array(errs_angle) <= 5))[0]) * 100. / (len(errs_trans) +
eps)
acc3d10 = len(np.where(
np.array(errs_3d) <= diam * 0.1)[0]) * 100. / (len(errs_3d) + eps)
acc5cm5deg = len(
np.where((np.array(errs_trans) <= 0.05)
& (np.array(errs_angle) <= 5))[0]) * 100. / (len(errs_trans) +
eps)
corner_acc = len(np.where(np.array(errs_corner2D) <= px_threshold)
[0]) * 100. / (len(errs_corner2D) + eps)
mean_err_2d = np.mean(errs_2d)
mean_corner_err_2d = np.mean(errs_corner2D)
nts = float(testing_samples)
if testtime:
print('-----------------------------------')
print(' tensor to cuda : %f' % (t2 - t1))
print(' predict : %f' % (t3 - t2))
print('get_region_boxes : %f' % (t4 - t3))
print(' eval : %f' % (t5 - t4))
print(' total : %f' % (t5 - t1))
print('-----------------------------------')
# Print test statistics
logging('Results of {}'.format(name))
logging(' Acc using {} px 2D Projection = {:.2f}%'.format(
px_threshold, acc))
logging(' Acc using 10% threshold - {} vx 3D Transformation = {:.2f}%'.
format(diam * 0.1, acc3d10))
logging(' Acc using 5 cm 5 degree metric = {:.2f}%'.format(acc5cm5deg))
logging(
" Mean 2D pixel error is %f, Mean vertex error is %f, mean corner error is %f"
% (mean_err_2d, np.mean(errs_3d), mean_corner_err_2d))
logging(
' Translation error: %f m, angle error: %f degree, pixel error: % f pix'
% (testing_error_trans / nts, testing_error_angle / nts,
testing_error_pixel / nts))
if save:
predfile = backupdir + '/predictions_linemod_' + name + '.mat'
scipy.io.savemat(
predfile, {
'R_gts': gts_rot,
't_gts': gts_trans,
'corner_gts': gts_corners2D,
'R_prs': preds_rot,
't_prs': preds_trans,
'corner_prs': preds_corners2D
})
def main(camera_id):
ip = str('192.168.0.2')
name = str('admin')
pw = str('a1234567')
camera = HKCamera(ip, name, pw)
threadPubMsg_shelfID_1 = pubmsg.MsgPublishClass(cameraID=camera_id,
shelfID=1)
threadPubMsg_shelfID_1.setDaemon(True)
threadPubMsg_shelfID_1.start()
threadPubMsg_shelfID_2 = pubmsg.MsgPublishClass(cameraID=camera_id,
shelfID=2)
threadPubMsg_shelfID_2.setDaemon(True)
threadPubMsg_shelfID_2.start()
threadPubMsg_dict = {
'shelfID_1': threadPubMsg_shelfID_1,
'shelfID_2': threadPubMsg_shelfID_2
}
model = loadDataset()
cfg = Darknet('cfg/yolov3.cfg')
cfg.load_weights('yolov3.weights')
cfg.cuda()
# global frame_number
frame_number2 = [0]
flag = [0]
bridge = CvBridge()
dic_change = {}
dic_change_huojia2 = {}
huojia1_id = 1
huojia2_id = 2
while not rospy.is_shutdown():
frame_origin = camera.getFrame()
frame_origin = np.array(frame_origin)
frame_origin = cv2.resize(frame_origin,
None,
fx=0.75,
fy=0.75,
interpolation=cv2.INTER_AREA)
frame_trans = copy.deepcopy(frame_origin)
# # draw the shangping area
# left_x, top_y, right_m, bottom_n = shangpin_area()
# cv2.rectangle(frame_origin, (left_x, top_y), (right_m, bottom_n), (0, 255, 0), 2)
#
# left_x_2, top_y_2, right_m_2, bottom_n_2 = shangpin_area_huojia2()
# cv2.rectangle(frame_origin, (left_x_2, top_y_2), (right_m_2, bottom_n_2), (255, 0, 0), 2)
res, camera_id = callback((None, cfg, model, frame_number2, bridge,
camera_id, flag, frame_origin))
if res == []:
threadPubMsg = threadPubMsg_dict['shelfID_' + str(huojia1_id)]
threadPubMsg.set_commodity_recognition_trigger_with_image(
camera_id=camera_id,
person_id=-1,
shelf_id=-1,
flag=0,
flag1=0,
flag2=0,
flag_list=[],
frame=frame_trans)
threadPubMsg = threadPubMsg_dict['shelfID_' + str(huojia2_id)]
threadPubMsg.set_commodity_recognition_trigger_with_image(
camera_id=camera_id,
person_id=-1,
shelf_id=-1,
flag=0,
flag1=0,
flag2=0,
flag_list=[],
frame=frame_trans)
continue
dic, dic_huojia2 = xuanze_original(res, frame_origin, model, cfg,
camera_id, dic_change,
dic_change_huojia2, huojia1_id,
huojia2_id)
if compare_dic(dic, dic_change) == False and compare_dic(
dic_huojia2, dic_change_huojia2) == False:
pass
else:
dic, dic_huojia2 = xuanze(res, frame_origin, model, cfg,
threadPubMsg_dict, camera_id, dic,
dic_change, dic_huojia2,
dic_change_huojia2, huojia1_id,
huojia2_id, frame_trans)
print("**********************")
print("dic_change shelf1: {}".format(dic))
print("dic_change_shelf2: {}".format(dic_huojia2))
print("")
dic_change = dic
dic_change_huojia2 = dic_huojia2
HKIPcamera.release()
init_epoch = model.seen/nsamples
kwargs = {'num_workers': num_workers, 'pin_memory': True} if use_cuda else {}
test_loader = torch.utils.data.DataLoader(
dataset.listDataset(testlist, shape=(init_width, init_height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]), train=False),
batch_size=batch_size, shuffle=False, **kwargs)
if use_cuda:
if ngpus > 1:
model = torch.nn.DataParallel(model).cuda()
else:
model = model.cuda()
params_dict = dict(model.named_parameters())
params = []
for key, value in params_dict.items():
if key.find('.bn') >= 0 or key.find('.bias') >= 0:
params += [{'params': [value], 'weight_decay': 0.0}]
else:
params += [{'params': [value], 'weight_decay': decay*batch_size}]
optimizer = optim.SGD(model.parameters(), lr=learning_rate/batch_size, momentum=momentum, dampening=0, weight_decay=decay*batch_size)
def adjust_learning_rate(optimizer, batch):
"""Sets the learning rate to the initial LR decayed by 10 every 30 epochs"""
lr = learning_rate
for i in range(len(steps)):
scale = scales[i] if i < len(scales) else 1
def generate_det(seq_dir, npy_dir, cfg_file, weight_file):
# load yolo model
m = Darknet(cfg_file)
m.print_network()
m.load_weights(weight_file)
print('Loading weights from %s... Done!' % (weight_file))
use_cuda = 1
if use_cuda:
m.cuda()
if not os.path.exists(npy_dir):
os.makedirs(npy_dir)
seq_list = glob.glob(os.path.join(seq_dir, "*-YOLO"))
seq_list = sorted(seq_list)
for seq in seq_list:
seq_name = os.path.basename(seq)
print("processing: %s" % seq_name)
det_dir = os.path.join(seq, "det")
if not os.path.exists(det_dir):
os.makedirs(det_dir)
txt_file = os.path.join(det_dir, "det.txt")
fid = open(txt_file, 'w')
npy_file = os.path.join(npy_dir, seq_name + ".npy")
img_dir = os.path.join(seq, "img1")
img_list = os.listdir(img_dir)
img_list = sorted(img_list)
img = cv2.imread(os.path.join(img_dir, img_list[0]))
sized = cv2.resize(img, (m.width, m.height))
sized = cv2.cvtColor(sized, cv2.COLOR_BGR2RGB)
boxes = do_detect(m, sized, 0.5, 0.4, use_cuda)
total_time = 0.0
npy_list = []
for i in range(len(img_list)):
print("processing: %d/%d" % (i + 1, len(img_list)))
img_name = img_list[i][:-4]
img_idx = int(img_name)
img_path = os.path.join(img_dir, img_list[i])
img = cv2.imread(img_path)
sized = cv2.resize(img, (m.width, m.height))
sized = cv2.cvtColor(sized, cv2.COLOR_BGR2RGB)
#time_0 = time.time()
boxes = do_detect(m, sized, 0.5, 0.4, use_cuda)
#time_1 = time.time()
#total_time += time_1 - time_0
height, width = img.shape[:2]
for j in range(len(boxes)):
box = boxes[j]
cls_id = box[6]
if cls_id != 0:
continue
x = (box[0] - box[2] / 2.0) * width
y = (box[1] - box[3] / 2.0) * height
w = box[2] * width
h = box[3] * height
cls_conf = box[5]
txt_tmp = "%d,-1,%.1f,%.1f,%.1f,%.1f,%.3f\n" % (img_idx, x, y,
w, h, cls_conf)
fid.write(txt_tmp)
npy_list.append([
img_idx, -1.0, x, y, w, h, cls_conf, -1.0, -1.0, -1.0, 1.0
])
fid.close()
np.save(npy_file,
np.asarray(npy_list, dtype=np.float32),
allow_pickle=False)
def valid(datacfg, cfgfile, weightfile, outfile):
options = read_data_cfg(datacfg)
valid_images = options['valid']
name_list = options['names']
prefix = 'results'
names = load_class_names(name_list)
with open(valid_images) as fp:
tmp_files = fp.readlines()
valid_files = [item.rstrip() for item in tmp_files]
m = Darknet(cfgfile)
m.print_network()
m.load_weights(weightfile)
m.cuda()
m.eval()
valid_dataset = dataset.listDataset(valid_images,
shape=(m.width, m.height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]))
valid_batchsize = 2
assert (valid_batchsize > 1)
kwargs = {'num_workers': 4, 'pin_memory': True}
valid_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=valid_batchsize,
shuffle=False,
**kwargs)
fps = [0] * m.num_classes
if not os.path.exists('results'):
os.mkdir('results')
for i in range(m.num_classes):
buf = '%s/%s%s.txt' % (prefix, outfile, names[i])
fps[i] = open(buf, 'w')
lineId = -1
conf_thresh = 0.005
nms_thresh = 0.45
for batch_idx, (data, target) in enumerate(valid_loader):
data = data.cuda()
data = Variable(data, volatile=True)
output = m(data).data
batch_boxes = get_region_boxes(output, conf_thresh, m.num_classes,
m.anchors, m.num_anchors, 0)
for i in range(output.size(0)):
lineId = lineId + 1
fileId = os.path.basename(valid_files[lineId]).split('.')[0]
width, height = get_image_size(valid_files[lineId])
print(valid_files[lineId])
boxes = batch_boxes[i]
boxes = nms(boxes, nms_thresh)
for box in boxes:
x1 = (box[0] - box[2] / 2.0) * width
y1 = (box[1] - box[3] / 2.0) * height
x2 = (box[0] + box[2] / 2.0) * width
y2 = (box[1] + box[3] / 2.0) * height
det_conf = box[4]
cls_conf = box[5]
cls_id = box[6]
prob = det_conf * cls_conf
fps[cls_id].write('%s %f %f %f %f %f\n' %
(fileId, prob, x1, y1, x2, y2))
for i in range(m.num_classes):
fps[i].close()
def main(camera_id, shelf_id):
rospy.init_node('MultiProcessingNode', anonymous=True)
ip = '192.168.0.' + str(camera_id)
name = str('admin')
pw = str('a1234567')
camera = HKCamera(ip, name, pw)
threadPubMsg_shelfID_1 = pubmsg.MsgPublishClass(cameraID=camera_id,
shelfID=shelf_id[0])
threadPubMsg_shelfID_1.setDaemon(True)
threadPubMsg_shelfID_1.start()
shelf1 = 'shelfID_' + str(shelf_id[0])
threadPubMsg_dict = {shelf1: threadPubMsg_shelfID_1}
model = loadDataset()
cfg = Darknet('cfg/yolov3.cfg')
cfg.load_weights('yolov3.weights')
cfg.cuda()
# global frame_number
frame_number2 = [0]
flag = [0]
bridge = CvBridge()
dic_change = {}
pre_res = {}
huojia1_id = shelf_id[0]
print("huojia1_id: {}".format(huojia1_id))
tmp = 0
while not rospy.is_shutdown():
frame_origin = camera.getFrame()
frame_origin = np.array(frame_origin)
frame_origin = cv2.resize(frame_origin,
None,
fx=0.75,
fy=0.75,
interpolation=cv2.INTER_AREA)
frame_trans = copy.deepcopy(frame_origin)
# draw the shangping area
# left_x, top_y, right_m, bottom_n = shangpin_area(huojia1_id)
# cv2.rectangle(frame_origin, (left_x, top_y), (right_m, bottom_n), (0, 255, 0), 2)
res, camera_id, dict_res = callback(
(None, cfg, model, frame_number2, bridge, camera_id, flag,
frame_origin, huojia1_id, pre_res))
if res == []:
if tmp > 30:
threadPubMsg = threadPubMsg_dict['shelfID_' + str(huojia1_id)]
threadPubMsg.set_commodity_recognition_trigger_with_image(
camera_id=camera_id,
person_id=-1,
shelf_id=huojia1_id,
flag=0,
flag1=0,
flag2=0,
flag_list=[],
frame=None)
tmp = 0
else:
tmp += 1
continue
else:
tmp = 0
dic = xuanze_original(res, frame_origin, model, cfg, camera_id,
dic_change, huojia1_id, pre_res)
if compare_dic(dic, dic_change) == False:
pass
else:
dic = xuanze(res, frame_origin, model, cfg, threadPubMsg_dict,
camera_id, dic, dic_change, huojia1_id, frame_trans,
pre_res)
#print("**********************")
#print("dic_change_shelf_{}: {}".format(shelf_id[0], dic))
#print("")
change_idnum = len(pre_res.keys()) == len(res)
if change_idnum:
pre_res = dict_res
else:
pre_res = {}
dic_change = dic
HKIPcamera.release()
class Detector:
def __init__(self, resolution=416):
'''
:param resolution: int, multiple of 32 greater than 32
'''
self.batch_size = 1
self.scales = [1, 2, 3]
self.resolution = resolution
self.num_boxes = [
self.resolution // 8, self.resolution // 16, self.resolution // 32
]
self.num_boxes = sum([3 * (x**2) for x in self.num_boxes])
self.scales_indices = []
for scale in self.scales:
li = list(
range((scale - 1) * self.num_boxes // 3,
scale * self.num_boxes // 3))
self.scales_indices.extend(li)
self.confidence = 0.5
self.nms_thresh = 0.4
self.start = 0
self.save_directory = '.'
self.cfg_file = 'cfg/yolov3.cfg'
self.weights_file = "yolov3.weights"
self.colors = pkl.load(open("pallete", "rb"))
self.CUDA = torch.cuda.is_available()
self.num_classes = 80
self.classes = load_classes('data/coco.names')
# Set up the neural network
print("Loading network.....")
self.model = Darknet(self.cfg_file)
self.model.load_weights(self.weights_file)
print("Network successfully loaded")
self.model.net_info["height"] = self.resolution
self.inp_dim = self.model.net_info["height"]
assert self.inp_dim % 32 == 0
assert self.inp_dim > 32
# If there's a GPU availible, put the model on GPU
if self.CUDA:
self.model.cuda()
# Set the model in evaluation mode
self.model.eval()
def detect_objects(self, image_path):
image_prep = prep_image(image_path, self.inp_dim)
im_batches = [image_prep[0]]
orig_ims = [image_prep[1]]
im_dim_list = [image_prep[2]]
im_dim_list = torch.FloatTensor(im_dim_list).repeat(1, 2)
img_path = image_path
if self.CUDA:
im_dim_list = im_dim_list.cuda()
write = False
self.model(get_test_input(self.inp_dim, self.CUDA), self.CUDA)
objs = {}
i = 0
for batch in im_batches:
if self.CUDA:
batch = batch.cuda()
with torch.no_grad():
prediction = self.model(Variable(batch), self.CUDA)
prediction = prediction[:, self.scales_indices]
prediction = write_results(prediction,
self.confidence,
self.num_classes,
nms=True,
nms_conf=self.nms_thresh)
prediction[:, 0] += i * self.batch_size
if not write:
output = prediction
write = 1
else:
output = torch.cat((output, prediction))
for im_num, image in enumerate(img_path[i * self.batch_size:min(
(i + 1) * self.batch_size, len(img_path))]):
im_id = i * self.batch_size + im_num
objs = [
self.classes[int(x[-1])] for x in output
if int(x[0]) == im_id
]
print("{0:20s} {1:s}".format("Objects Detected:",
" ".join(objs)))
print(
"----------------------------------------------------------"
)
i += 1
if self.CUDA:
torch.cuda.synchronize()
try:
output
except NameError:
print("No detections were made")
exit()
im_dim_list = torch.index_select(im_dim_list, 0, output[:, 0].long())
scaling_factor = torch.min(self.inp_dim / im_dim_list,
1)[0].view(-1, 1)
output[:,
[1, 3]] -= (self.inp_dim -
scaling_factor * im_dim_list[:, 0].view(-1, 1)) / 2
output[:,
[2, 4]] -= (self.inp_dim -
scaling_factor * im_dim_list[:, 1].view(-1, 1)) / 2
output[:, 1:5] /= scaling_factor
for i in range(output.shape[0]):
output[i, [1, 3]] = torch.clamp(output[i, [1, 3]], 0.0,
im_dim_list[i, 0])
output[i, [2, 4]] = torch.clamp(output[i, [2, 4]], 0.0,
im_dim_list[i, 1])
def write(x, batches, results):
c1 = tuple(x[1:3].int())
c2 = tuple(x[3:5].int())
img = results[int(x[0])]
cls = int(x[-1])
label = "{0}".format(self.classes[cls])
color = random.choice(self.colors)
cv2.rectangle(img, c1, c2, color, 1)
t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 1, 1)[0]
c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4
cv2.rectangle(img, c1, c2, color, -1)
cv2.putText(img, label, (c1[0], c1[1] + t_size[1] + 4),
cv2.FONT_HERSHEY_PLAIN, 1, [225, 255, 255], 1)
return img
list(map(lambda x: write(x, im_batches, orig_ims), output))
det_names = pd.Series(img_path).apply(
lambda x: "{}/det_{}".format(self.save_directory,
x.split("/")[-1]))
cv2.imwrite(det_names[0], orig_ims[0])
torch.cuda.empty_cache()
ret_path = det_names[0]
return ret_path, objs, orig_ims[0]
def evaluate_with_gt_pos(cfgfile,
weightfile,
listfile,
append,
bestCnt,
withZoom=True,
use_cuda=True,
zoom_type=None):
import cv2
all_channels = [
32, 64, 32, 64, 128, 64, 128, 64, 128, 256, 128, 256, 128, 256, 128,
256, 128, 256, 128, 256, 128, 256, 128, 256, 128, 256, 512, 256, 512,
256, 512, 256, 512, 256, 512, 256, 512, 256, 512, 256, 512, 256, 512,
1024, 512, 1024, 512, 1024, 512, 1024, 512, 1024, 512, 1024, 512, 1024,
512, 256, 256, 512, 256, 512, 256, 128, 128, 256, 128, 256, 128, 256,
2, 512, 1024, 512, 1024, 512, 256, 256, 512, 256, 512, 256, 128, 128,
256, 128, 256, 128, 256, 24
]
m = Darknet(cfgfile, all_channels)
m.print_network()
m.load_state_dict(torch.load(weightfile))
print('Loading weights from %s... Done!' % (weightfile))
if use_cuda:
m.cuda()
m.print_bn_weights()
with open(listfile, 'r') as file:
imglines = file.readlines()
failed_pred = 0
total_pred = 0
for idx in range(len(imglines)):
max_conf = 0
imgfile = imglines[idx].rstrip()
img = cv2.imread(imgfile)
dirname, filename = os.path.split(imgfile)
baseName, _ = os.path.splitext(filename)
dirname = os.path.splitext(dirname[dirname.rfind('/') + 1:])[0]
outFileName = dirname + '_' + baseName
start = time.time()
gtPoses = [None] * 3
rawk = K_tango
target_shape = (704, 704)
max_conf = 0
best_pred = None
best_border = None
save = False
print('imgfile', imgfile)
print(str(failed_pred) + ' ' + str(total_pred))
#predPose, conf, deviation, p2d = do_detect(m, img, rawk, gtPoses, bestCnt, 0, 0, use_cuda)
try:
total_pred = total_pred + 1
predPose, conf, p2d = do_detect(m, img, rawk, gtPoses, bestCnt, 0,
0, use_cuda)
except Exception:
failed_pred = failed_pred + 1
pass
finish = time.time()
name = 'img/' + filename + '.png'
if predPose is not None and (len(predPose) != 0):
pose = predPose[0][1]
print(str(pose))
r = pose[:, 3]
name = 'img/' + filename + '_' + str(conf) + '.png'
#save_img_with_label(img, pose, rawk, name)
print(name)
quat = np.delete(pose, 3, axis=1).T
q0, qvec = dcm2quat(quat)
q = [q0, qvec[0], qvec[1], qvec[2]]
print(conf)
else:
name = 'img/missing' + filename + '.png'
print('problem', name, save)
print(str(failed_pred) + ' ' + str(total_pred))
class Main(QtWidgets.QMainWindow, Ui_MainWindow):
logQueue = multiprocessing.Queue() # 日志数据队列，用于多进程之间传输数据
receiveLogSignal = pyqtSignal(str)
def __init__(self):
QtWidgets.QMainWindow.__init__(self)
self.setupUi(self)
self.cap = None
# video
self.center()
self.openFIleButton.clicked.connect(self.open_video)
self.closeFileButton.clicked.connect(self.close_video)
# 创建一个关闭事件并设为未触发
self.stopEvent = threading.Event()
self.stopEvent.clear()
# 加载模型
self.load_models.clicked.connect(self.load_model)
# 加载日志
self.receiveLogSignal.connect(lambda log: self.logOutput(log))
self.logOutputThread = threading.Thread(target=self.receiveLog,
daemon=True)
self.logOutputThread.start()
# 调节帧率
self.changeFrameSlider.valueChanged.connect(self.frameChange)
self.frameInterval = self.changeFrameSlider.value()
def frameChange(self):
self.label_14.setText(str(self.changeFrameSlider.value()))
#print("frameInterval:" + str(self.frameInterval))
def logOutput(self, log):
# 获取当前系统时间
time = datetime.now().strftime('[%Y/%m/%d %H:%M:%S]')
log = time + '\n' + log
self.logFile.write(log)
self.textEdit.moveCursor(QTextCursor.End)
self.textEdit.insertPlainText(log)
self.textEdit.ensureCursorVisible() # 自动滚屏
def receiveLog(self):
while True:
data = self.logQueue.get()
if data:
self.receiveLogSignal.emit(data)
else:
continue
def center(self, screenNum=0):
screen = QDesktopWidget().screenGeometry()
size = self.geometry()
self.normalGeometry2 = QRect(
(screen.width() - size.width()) / 2 + screen.left(),
(screen.height() - size.height()) / 2, size.width(), size.height())
self.setGeometry((screen.width() - size.width()) / 2 + screen.left(),
(screen.height() - size.height()) / 2, size.width(),
size.height())
def open_video(self):
fileName, _ = QFileDialog.getOpenFileName(self, "载入监控视频", '../videos')
self.cap = cv2.VideoCapture(fileName)
self.frameRate = self.cap.get(cv2.CAP_PROP_FPS)
video_thread = threading.Thread(target=self.display_video)
video_thread.start()
self.logFile = open('../log/log_info.txt', 'a')
def close_video(self):
self.stopEvent.set()
def display_video(self):
self.openFIleButton.setEnabled(False)
self.closeFileButton.setEnabled(True)
# RGB转BGR
frames = 0
while self.cap.isOpened():
ret, frame = self.cap.read()
if ret:
if frames % self.changeFrameSlider.value() == 0:
#frame = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
plate_frame = frame.copy()
img = frame.copy()
output = None
orign_img = None
# 系统日志
count_info_log = ""
event_info_log = ""
break_info_log = ""
try:
if self.target_detect.isChecked(): # 目标识别
output, orign_img, img, pedestrians_num = target_detect(
self.model, frame)
if int(pedestrians_num) != 0:
break_info_log = str(
pedestrians_num) + "人闯红灯;\n"
self.break_traffic_warning.setVisible(True)
self.break_traffic_label.setVisible(True)
self.break_traffic_label.setText(
break_info_log)
else:
self.break_traffic_label.setVisible(False)
self.break_traffic_warning.setVisible(False)
# 红绿灯检测
if self.traffic_light_detect.isChecked():
traffic_light_color = traffic_light_detect(
output, orign_img)
if traffic_light_color == "green":
self.red_light.setVisible(False)
self.green_light.setVisible(True)
elif traffic_light_color == "red":
self.green_light.setVisible(False)
self.red_light.setVisible(True)
else:
self.green_light.setVisible(False)
self.red_light.setVisible(False)
else:
self.green_light.setVisible(False)
self.red_light.setVisible(False)
#车流，人流检测
people_num = 0
cars_num = 0
motors_num = 0
if self.cars_detect.isChecked():
_, cars_num, motors_num = classNum_detect(output)
self.tableWidget.setItem(
0, 1, QTableWidgetItem(str(cars_num)))
self.tableWidget.setItem(
0, 2, QTableWidgetItem(str(motors_num)))
else:
self.tableWidget.setItem(0, 1,
QTableWidgetItem(str(0)))
self.tableWidget.setItem(0, 2,
QTableWidgetItem(str(0)))
#print(1111)
if self.people_detect.isChecked():
people_num, _, _ = classNum_detect(output)
self.tableWidget.setItem(
0, 0, QTableWidgetItem(str(people_num)))
else:
self.tableWidget.setItem(0, 0,
QTableWidgetItem(str(0)))
count_info_log = "people:" + str(people_num) + ", cars:" + str(cars_num) + \
", motors:" + str(motors_num) + ";\n"
# 车牌识别
if self.license_plate_detect.isChecked():
plate_info_list = recognize_plate(plate_frame)
for plate_info in plate_info_list:
plate = plate_info[0] # 车牌
conficdence = plate_info[1] # 置信度
#print("车牌:" + plate)
#self.license_result.clear()
self.license_result.setText(plate)
rect = plate_info[2] # 位置
#print(rect[0], rect[2], rect[1], rect[3])
plate_img = plate_frame[int(rect[1]) : int(rect[3] + rect[1]), \
int(rect[0]) : int(rect[2]+rect[0])]
plate_img = cv2.cvtColor(
plate_img, cv2.COLOR_RGB2BGR)
plate_img = cv2.resize(plate_img, (140, 30))
plate_img = QImage(plate_img.data, plate_img.shape[1], \
plate_img.shape[0], QImage.Format_RGB888)
self.license_graph.setPixmap(
QPixmap.fromImage(plate_img))
img = drawRectBox(img, rect, plate)
plate_center = [
int(rect[0] + rect[2]),
int(rect[1] + rect[3])
] #车牌中心中心位置
car_color = detect_car_color(
output, plate_frame, plate_center) #检测汽车颜色
event_info_log = car_color + "汽车,车牌信息：" + plate + \
"识别准确率:" + str(conficdence)[:5] + '\n'
#print(event_info_log)
else:
self.license_graph.clear()
self.license_result.clear()
except:
pass
log_info = count_info_log + event_info_log + break_info_log
self.logQueue.put(log_info)
#self.sys_log(count_info_log, event_info_log, break_info_log)
#self.set_log_info(count_info_log, event_info_log, break_info_log)
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
img = cv2.resize(img, (1080, 540))
img = QImage(img.data, img.shape[1], img.shape[0],
QImage.Format_RGB888)
self.video_plate.setPixmap(QPixmap.fromImage(img))
cv2.waitKey(1)
frames += 1
#print(frames)
if self.stopEvent.is_set():
self.stopEvent.clear()
#self.textEdit.clear()
self.video_plate.clear()
self.tableWidget.setItem(0, 0, QTableWidgetItem(str(0)))
self.tableWidget.setItem(0, 1, QTableWidgetItem(str(0)))
self.tableWidget.setItem(0, 2, QTableWidgetItem(str(0)))
break
else:
self.video_plate.clear()
break
try:
self.openFIleButton.setEnabled(True)
self.cap.release()
self.logFile.close()
self.green_light.setVisible(False)
self.red_light.setVisible(False)
self.break_traffic_warning.setVisible(False)
self.break_traffic_label.setVisible(False)
self.license_graph.clear()
self.license_result.clear()
except:
print("资源释放错误")
def load_model(self):
CUDA = torch.cuda.is_available()
print("Loading network.....")
self.model = Darknet("../yolov3/cfg/yolov3.cfg")
self.model.load_weights("../yolov3/weights/yolov3.weights")
print("Network successfully loaded")
self.model.net_info["height"] = 416
inp_dim = int(self.model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
if CUDA:
self.model.cuda() # 将模型迁移到GPU
self.model.eval()
load_completed = QMessageBox.information(self, 'message', '模型加载完成．',
QMessageBox.Ok)
class YOLO:
def __init__(self,
cfg_file: pathlib.Path,
weights_file: pathlib.Path,
class_names_file: pathlib.Path,
resolution: int = 416,
class_filters: List[str] = None) -> None:
self.net: Any = None
self.input_dim: int = None
self.load_net(cfg_file, weights_file, resolution)
self.class_names = load_classes(class_names_file)
self.num_classes = len(self.class_names)
self.class_filters = class_filters
def load_net(self, cfg_file: pathlib.Path, weights_file: pathlib.Path,
resolution: int) -> None:
self.net = Darknet(str(cfg_file))
self.net.load_weights(str(weights_file))
self.net.net_info['height'] = resolution
self.net.cuda()
self.input_dim = self.net.net_info['height']
if self.input_dim % 32 != 0 or self.input_dim <= 32:
raise ValueError("Bad input dimension. Resolution is bad")
# self.net(get_test_input(self.input_dim, True), True)
self.net.eval()
def prep_frame(
self, frame: np.ndarray
) -> Tuple[np.ndarray, np.ndarray, Tuple[int, int]]:
original_frame = frame
dim = original_frame.shape[1], original_frame.shape[0]
frame = (letterbox_image(original_frame,
(self.input_dim, self.input_dim)))
frame_ = frame[:, :, ::-1].transpose((2, 0, 1)).copy()
frame_ = torch.from_numpy(frame_).float().div(255.0).unsqueeze(0)
return frame_, original_frame, dim
def format_output(self, output: Any, threshold: float,
frame_dimensions: Tuple[int, int]) -> Optional[Any]:
output = write_results(output,
threshold,
self.num_classes,
nms=True,
nms_conf=threshold)
if isinstance(output, int):
# means no output
return None
frame_dimensions = frame_dimensions.repeat(output.size(0), 1)
scaling_factor = torch.min(self.input_dim / frame_dimensions,
1)[0].view(-1, 1)
output[:, [1, 3]] -= (self.input_dim - scaling_factor *
frame_dimensions[:, 0].view(-1, 1)) / 2
output[:, [2, 4]] -= (self.input_dim - scaling_factor *
frame_dimensions[:, 1].view(-1, 1)) / 2
output[:, 1:5] /= scaling_factor
for i in range(output.shape[0]):
output[i, [1, 3]] = torch.clamp(output[i, [1, 3]], 0.0,
frame_dimensions[i, 0])
output[i, [2, 4]] = torch.clamp(output[i, [2, 4]], 0.0,
frame_dimensions[i, 1])
return output
def get_detections(self,
frame: np.ndarray,
threshold: float = 0.7) -> FrameAnnotations:
if frame is None:
return FrameAnnotations(frame=frame,
objects=list(),
image_width=None,
image_height=None)
new_frame, frame, dimensions = self.prep_frame(frame)
new_frame = new_frame.cuda()
frame_dimensions = torch.FloatTensor(dimensions).repeat(1, 2).cuda()
with torch.no_grad():
output = self.net(Variable(new_frame), True)
output = self.format_output(output, threshold, frame_dimensions)
objects = list()
if output is not None:
for obj in output:
if self.class_filters is not None:
if self.class_names[int(obj[-1])] not in \
self.class_filters:
continue
objects.append(
Object(class_name=self.class_names[int(obj[-1])],
bbox=BBox(left=int(obj[1]),
top=int(obj[2]),
right=int(obj[3]),
bottom=int(obj[4]))))
return FrameAnnotations(frame=frame,
objects=objects,
image_width=frame.shape[1],
image_height=frame.shape[0])
def main():
global args, best_prec1
args = parser.parse_args()
# create model
if args.pretrained:
print("=> using pre-trained model '{}'".format(args.arch))
# if args.arch == 'resnet50':
# import resnet_model
# model = resnet_model.resnet50_new(pretrained=True)
# print('save resnet50 to resnet50.weights')
# model.saveas_darknet_weights('resnet50.weights')
if args.arch == 'resnet50-darknet':
from darknet import Darknet
model = Darknet('cfg/resnet50.cfg')
print('load weights from resnet50.weights')
model.load_weights('resnet50.weights')
elif args.arch == 'resnet50-kaiming':
from caffenet import CaffeNet
model = CaffeNet('ResNet-50-deploy.prototxt')
print('load weights from ResNet-50-model.caffemodel')
model.load_weights('ResNet-50-model.caffemodel')
elif args.arch == 'resnet50-kaiming-dk':
from darknet import Darknet
model = Darknet('ResNet-50-model.cfg')
print('load weights from ResNet-50-model.weights')
model.load_weights('ResNet-50-model.weights')
elif args.arch == 'resnet18-caffe':
from caffenet import CaffeNet
model = CaffeNet('cfg/resnet-18.prototxt')
print('load weights from resnet-18.caffemodel')
model.load_weights('resnet-18.caffemodel')
elif args.arch == 'resnet18-darknet':
from darknet import Darknet
model = Darknet('resnet-18.cfg')
print('load weights from resnet-18.weights')
model.load_weights('resnet-18.weights')
elif args.arch == 'resnet50-test':
from darknet import Darknet
model = Darknet('test/ResNet-50-model.cfg')
print('load weights from test/ResNet-50-model.weights')
model.load_weights('test/ResNet-50-model.weights')
else:
model = models.__dict__[args.arch](pretrained=True)
else:
print("=> creating model '{}'".format(args.arch))
if args.arch.startswith('mobilenet'):
model = Net()
print(model)
else:
model = models.__dict__[args.arch]()
if args.arch.startswith('alexnet') or args.arch.startswith('vgg'):
model.features = torch.nn.DataParallel(model.features)
model.cuda()
else:
model = torch.nn.DataParallel(model).cuda()
# define loss function (criterion) and optimizer
criterion = nn.CrossEntropyLoss().cuda()
optimizer = torch.optim.SGD(model.parameters(),
args.lr,
momentum=args.momentum,
weight_decay=args.weight_decay)
# optionally resume from a checkpoint
if args.resume:
if os.path.isfile(args.resume):
print("=> loading checkpoint '{}'".format(args.resume))
checkpoint = torch.load(args.resume)
args.start_epoch = checkpoint['epoch']
best_prec1 = checkpoint['best_prec1']
model.load_state_dict(checkpoint['state_dict'])
optimizer.load_state_dict(checkpoint['optimizer'])
print("=> loaded checkpoint '{}' (epoch {})".format(
args.resume, checkpoint['epoch']))
else:
print("=> no checkpoint found at '{}'".format(args.resume))
cudnn.benchmark = True
# Data loading code
traindir = os.path.join(args.data, 'train')
valdir = os.path.join(args.data, 'val')
if args.arch == 'resnet50-test' or args.arch == 'resnet50-kaiming' or args.arch == 'resnet50-kaiming-dk':
normalize = transforms.Normalize(mean=[0.0, 0.0, 0.0],
std=[1.0, 1.0, 1.0])
elif args.arch == 'resnet18-darknet' or args.arch == 'resnet18-caffe':
normalize = transforms.Normalize(
mean=[104 / 255.0, 117 / 255.0, 123 / 255.0], std=[1.0, 1.0, 1.0])
else:
normalize = transforms.Normalize(mean=[0.485, 0.456, 0.406],
std=[0.229, 0.224, 0.225])
train_loader = torch.utils.data.DataLoader(datasets.ImageFolder(
traindir,
transforms.Compose([
transforms.RandomSizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
normalize,
])),
batch_size=args.batch_size,
shuffle=True,
num_workers=args.workers,
pin_memory=True)
val_loader = torch.utils.data.DataLoader(datasets.ImageFolder(
valdir,
transforms.Compose([
transforms.Scale(256),
transforms.CenterCrop(224),
transforms.ToTensor(),
normalize,
])),
batch_size=args.batch_size,
shuffle=False,
num_workers=args.workers,
pin_memory=True)
if args.evaluate:
validate(val_loader, model, criterion)
return
for epoch in range(args.start_epoch, args.epochs):
adjust_learning_rate(optimizer, epoch)
# train for one epoch
train(train_loader, model, criterion, optimizer, epoch)
# evaluate on validation set
prec1 = validate(val_loader, model, criterion)
# remember best prec@1 and save checkpoint
is_best = prec1 > best_prec1
best_prec1 = max(prec1, best_prec1)
save_checkpoint(
{
'epoch': epoch + 1,
'arch': args.arch,
'state_dict': model.state_dict(),
'best_prec1': best_prec1,
'optimizer': optimizer.state_dict(),
}, is_best)
class Detector:
def __init__(self, showTags=False, showCordenates=False):
self.showTags = showTags
self.showCordenates = showCordenates
self.model = Darknet(cfgfile)
self.model.load_weights(weightsfile)
self.model.net_info["height"] = resolution
self.inp_dim = int(self.model.net_info["height"])
assert self.inp_dim % 32 == 0
assert self.inp_dim > 32
# If there's a GPU availible, put the model on GPU
if CUDA:
self.model.cuda()
self.model.eval()
self.reconocimiento_facil = ReconocimientoFacial()
def detect(self, frame, debugFrames=[], frame_counter=0):
img = prep_image(frame, self.inp_dim)
im_dim = frame.shape[1], frame.shape[0]
im_dim = torch.FloatTensor(im_dim).repeat(1, 2)
if CUDA:
im_dim = im_dim.cuda()
img = img.cuda()
with torch.no_grad():
# output = self.model(Variable(img, volatile=True), CUDA)
output = self.model(Variable(img), CUDA)
output = write_results(output,
confidence,
num_classes,
nms_conf=nms_thesh)
if isinstance(output, int):
return frame
im_dim = im_dim.repeat(output.size(0), 1)
scaling_factor = torch.min(416 / im_dim, 1)[0].view(-1, 1)
output[:, [1, 3]] -= (self.inp_dim -
scaling_factor * im_dim[:, 0].view(-1, 1)) / 2
output[:, [2, 4]] -= (self.inp_dim -
scaling_factor * im_dim[:, 1].view(-1, 1)) / 2
output[:, 1:5] /= scaling_factor
for i in range(output.shape[0]):
output[i, [1, 3]] = torch.clamp(output[i, [1, 3]], 0.0, im_dim[i,
0])
output[i, [2, 4]] = torch.clamp(output[i, [2, 4]], 0.0, im_dim[i,
1])
if (self.showCordenates):
frame = lineaGol.add_linea_gol(frame, frame_counter)
list(
map(
lambda x: write(x, frame, self.showTags, self.
reconocimiento_facil), output))
return frame
CUDA = torch.cuda.is_available() # cuda가 사용가능한 상황인지
num_classes = 80 # 암튼 80
bbox_attrs = 5 + num_classes # Bouding Box 속성
model = Darknet(cfgfile) # Darknet
model.load_weights(weightsfile) # Model에 weighs파일을 load해준다
model.net_info["height"] = args.reso
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
if CUDA:
model.cuda() # Cuda를 사용중이면 model.cuda()
model.eval() # 모델 평가?
# cap = cv2.VideoCapture(0) #videoCapture(0) >> video 캡쳐변수 선언
cap = cv2.VideoCapture("http://192.168.0.54:8409/?action=snapshot")
# videoCapture("주소") >> video 캡쳐변수 선언
assert cap.isOpened(), 'Cannot capture source'
# assert는 가정설정문, 뒤의 조건이 True가 아니면 AssertError를 발생시킨다.
frames = 0
# frame 변수 선언, 초기값은 0
start = time.time() # 시간을 측정해주는 함수
while cap.isOpened(): # cap이 초기화가 잘 되어 있는지 확인
stdoutToServer=True,
stderrToServer=True)
datacfg = 'cfg/voc.data'
cfgfile = 'cfg/yolov2-tiny-voc.cfg'
weightfile = '../yolov2-tiny-bnn/weights/yolov2-tiny-voc.weights'
options = read_data_cfg(datacfg)
valid_images = options['valid']
with open(valid_images) as fp:
tmp_files = fp.readlines()
valid_files = [item.rstrip() for item in tmp_files]
m = Darknet(cfgfile)
m.print_network()
m.load_weights(weightfile)
m.cuda()
m.eval()
valid_dataset = dataset.listDataset(valid_images,
shape=(m.width, m.height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]))
valid_batchsize = 2
assert (valid_batchsize > 1)
kwargs = {'num_workers': 0, 'pin_memory': True}
valid_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=valid_batchsize,
shuffle=False,
def valid(datacfg, modelcfg, weightfile):
def truths_length(truths, max_num_gt=50):
for i in range(max_num_gt):
if truths[i][1] == 0:
return i
# Parse configuration files
data_options = read_data_cfg(datacfg)
valid_images = data_options['valid']
meshname = data_options['mesh']
backupdir = data_options['backup']
name = data_options['name']
gpus = data_options['gpus']
fx = float(data_options['fx'])
fy = float(data_options['fy'])
u0 = float(data_options['u0'])
v0 = float(data_options['v0'])
im_width = int(data_options['width'])
im_height = int(data_options['height'])
if not os.path.exists(backupdir):
makedirs(backupdir)
# Parameters
seed = int(time.time())
os.environ['CUDA_VISIBLE_DEVICES'] = gpus
torch.cuda.manual_seed(seed)
save = False
testtime = True
num_classes = 1
testing_samples = 0.0
if save:
makedirs(backupdir + '/test')
makedirs(backupdir + '/test/gt')
makedirs(backupdir + '/test/pr')
# To save
testing_error_trans = 0.0
testing_error_angle = 0.0
testing_error_pixel = 0.0
errs_2d = []
errs_3d = []
errs_trans = []
errs_angle = []
errs_corner2D = []
preds_trans = []
preds_rot = []
preds_corners2D = []
gts_trans = []
gts_rot = []
gts_corners2D = []
# Read object model information, get 3D bounding box corners
mesh = MeshPly(meshname)
vertices = np.c_[np.array(mesh.vertices),
np.ones((len(mesh.vertices), 1))].transpose()
corners3D = get_3D_corners(vertices)
try:
diam = float(options['diam'])
except:
diam = calc_pts_diameter(np.array(mesh.vertices))
# Read intrinsic camera parameters
intrinsic_calibration = get_camera_intrinsic(u0, v0, fx, fy)
# Get validation file names
with open(valid_images) as fp:
tmp_files = fp.readlines()
valid_files = [item.rstrip() for item in tmp_files]
# Specicy model, load pretrained weights, pass to GPU and set the module in evaluation mode
model = Darknet(modelcfg)
model.print_network()
model.load_weights(weightfile)
model.cuda()
model.eval()
test_width = model.test_width
test_height = model.test_height
num_keypoints = model.num_keypoints
num_labels = num_keypoints * 2 + 3 # +2 for width, height, +1 for class label
# Get the parser for the test dataset
valid_dataset = dataset.listDataset(valid_images,
shape=(test_width, test_height),
shuffle=False,
transform=transforms.Compose([
transforms.ToTensor(),
]))
# Specify the number of workers for multiple processing, get the dataloader for the test dataset
kwargs = {'num_workers': 4, 'pin_memory': True}
test_loader = torch.utils.data.DataLoader(valid_dataset,
batch_size=1,
shuffle=False,
**kwargs)
logging(" Testing {}...".format(name))
logging(" Number of test samples: %d" % len(test_loader.dataset))
# Iterate through test batches (Batch size for test data is 1)
count = 0
for batch_idx, (data, target) in enumerate(test_loader):
t1 = time.time()
# Pass data to GPU
# import IPython; IPython.embed()
data = data.cuda()
# target = target.cuda()
# Wrap tensors in Variable class, set volatile=True for inference mode and to use minimal memory during inference
data = Variable(data, volatile=True)
t2 = time.time()
# Forward pass
output = model(data).data
t3 = time.time()
# Using confidence threshold, eliminate low-confidence predictions
all_boxes = get_region_boxes(output, num_classes, num_keypoints)
all_boxes = [t.cpu() for t in all_boxes]
t4 = time.time()
# Evaluation
# Iterate through all batch elements
for box_pr, target in zip([all_boxes], [target[0]]):
# For each image, get all the targets (for multiple object pose estimation, there might be more than 1 target per image)
truths = target.view(-1, num_labels)
# Get how many objects are present in the scene
num_gts = truths_length(truths)
# Iterate through each ground-truth object
for k in range(num_gts):
box_gt = list()
for j in range(1, 2 * num_keypoints + 1):
box_gt.append(truths[k][j])
box_gt.extend([1.0, 1.0])
box_gt.append(truths[k][0])
# Denormalize the corner predictions
corners2D_gt = np.array(np.reshape(box_gt[:18], [-1, 2]),
dtype='float32')
corners2D_pr = np.array(np.reshape(box_pr[:18], [-1, 2]),
dtype='float32')
corners2D_gt[:, 0] = corners2D_gt[:, 0] * im_width
corners2D_gt[:, 1] = corners2D_gt[:, 1] * im_height
corners2D_pr[:, 0] = corners2D_pr[:, 0] * im_width
corners2D_pr[:, 1] = corners2D_pr[:, 1] * im_height
preds_corners2D.append(corners2D_pr)
gts_corners2D.append(corners2D_gt)
# Compute corner prediction error
corner_norm = np.linalg.norm(corners2D_gt - corners2D_pr,
axis=1)
corner_dist = np.mean(corner_norm)
errs_corner2D.append(corner_dist)
# Compute [R|t] by pnp
R_gt, t_gt = pnp(
np.array(np.transpose(
np.concatenate((np.zeros((3, 1)), corners3D[:3, :]),
axis=1)),
dtype='float32'), corners2D_gt,
np.array(intrinsic_calibration, dtype='float32'))
R_pr, t_pr = pnp(
np.array(np.transpose(
np.concatenate((np.zeros((3, 1)), corners3D[:3, :]),
axis=1)),
dtype='float32'), corners2D_pr,
np.array(intrinsic_calibration, dtype='float32'))
# Compute translation error
trans_dist = np.sqrt(np.sum(np.square(t_gt - t_pr)))
errs_trans.append(trans_dist)
# Compute angle error
angle_dist = calcAngularDistance(R_gt, R_pr)
errs_angle.append(angle_dist)
# Compute pixel error
Rt_gt = np.concatenate((R_gt, t_gt), axis=1)
Rt_pr = np.concatenate((R_pr, t_pr), axis=1)
proj_2d_gt = compute_projection(vertices, Rt_gt,
intrinsic_calibration)
proj_2d_pred = compute_projection(vertices, Rt_pr,
intrinsic_calibration)
norm = np.linalg.norm(proj_2d_gt - proj_2d_pred, axis=0)
pixel_dist = np.mean(norm)
errs_2d.append(pixel_dist)
# Compute 3D distances
transform_3d_gt = compute_transformation(vertices, Rt_gt)
transform_3d_pred = compute_transformation(vertices, Rt_pr)
norm3d = np.linalg.norm(transform_3d_gt - transform_3d_pred,
axis=0)
vertex_dist = np.mean(norm3d)
errs_3d.append(vertex_dist)
# Sum errors
testing_error_trans += trans_dist
testing_error_angle += angle_dist
testing_error_pixel += pixel_dist
testing_samples += 1
count = count + 1
if save:
preds_trans.append(t_pr)
gts_trans.append(t_gt)
preds_rot.append(R_pr)
gts_rot.append(R_gt)
np.savetxt(
backupdir + '/test/gt/R_' + valid_files[count][-8:-3] +
'txt', np.array(R_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/gt/t_' + valid_files[count][-8:-3] +
'txt', np.array(t_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/R_' + valid_files[count][-8:-3] +
'txt', np.array(R_pr, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/t_' + valid_files[count][-8:-3] +
'txt', np.array(t_pr, dtype='float32'))
np.savetxt(
backupdir + '/test/gt/corners_' +
valid_files[count][-8:-3] + 'txt',
np.array(corners2D_gt, dtype='float32'))
np.savetxt(
backupdir + '/test/pr/corners_' +
valid_files[count][-8:-3] + 'txt',
np.array(corners2D_pr, dtype='float32'))
t5 = time.time()
# Compute 2D projection error, 6D pose error, 5cm5degree error
px_threshold = 5 # 5 pixel threshold for 2D reprojection error is standard in recent sota 6D object pose estimation works
eps = 1e-5
acc = len(np.where(
np.array(errs_2d) <= px_threshold)[0]) * 100. / (len(errs_2d) + eps)
acc5cm5deg = len(
np.where((np.array(errs_trans) <= 0.05)
& (np.array(errs_angle) <= 5))[0]) * 100. / (len(errs_trans) +
eps)
acc3d10 = len(np.where(
np.array(errs_3d) <= diam * 0.1)[0]) * 100. / (len(errs_3d) + eps)
acc5cm5deg = len(
np.where((np.array(errs_trans) <= 0.05)
& (np.array(errs_angle) <= 5))[0]) * 100. / (len(errs_trans) +
eps)
corner_acc = len(np.where(np.array(errs_corner2D) <= px_threshold)
[0]) * 100. / (len(errs_corner2D) + eps)
mean_err_2d = np.mean(errs_2d)
mean_corner_err_2d = np.mean(errs_corner2D)
nts = float(testing_samples)
if testtime:
print('-----------------------------------')
print(' tensor to cuda : %f' % (t2 - t1))
print(' forward pass : %f' % (t3 - t2))
print('get_region_boxes : %f' % (t4 - t3))
print(' prediction time : %f' % (t4 - t1))
print(' eval : %f' % (t5 - t4))
print('-----------------------------------')
# Print test statistics
logging('Results of {}'.format(name))
logging(' Acc using {} px 2D Projection = {:.2f}%'.format(
px_threshold, acc))
logging(' Acc using 10% threshold - {} vx 3D Transformation = {:.2f}%'.
format(diam * 0.1, acc3d10))
logging(' Acc using 5 cm 5 degree metric = {:.2f}%'.format(acc5cm5deg))
logging(
" Mean 2D pixel error is %f, Mean vertex error is %f, mean corner error is %f"
% (mean_err_2d, np.mean(errs_3d), mean_corner_err_2d))
logging(
' Translation error: %f m, angle error: %f degree, pixel error: % f pix'
% (testing_error_trans / nts, testing_error_angle / nts,
testing_error_pixel / nts))
if save:
predfile = backupdir + '/predictions_linemod_' + name + '.mat'
scipy.io.savemat(
predfile, {
'R_gts': gts_rot,
't_gts': gts_trans,
'corner_gts': gts_corners2D,
'R_prs': preds_rot,
't_prs': preds_trans,
'corner_prs': preds_corners2D
})
'cautery': 'cfg/my_config_webcam.yaml'
} #
weightfile = {'hands': 'backup/hands/000500.weights'}
namesfile = {'hands': 'data/hands.names'}
#######################################################
# Setting up YOLO-hand
#######################################################
model_hand = Darknet(cfgfile['hands'])
model_hand.load_weights(weightfile['hands'])
print('Loading weights from %s... Done!' % (weightfile['hands']))
if use_cuda:
model_hand.cuda()
class_names = uyolo.load_class_names(namesfile['hands'])
#######################################################
# Setting up DOPE
#######################################################
yaml_path = cfgfile['cautery']
with open(yaml_path, 'r') as stream:
try:
print("Loading DOPE parameters from '{}'...".format(yaml_path))
params = yaml.load(stream)
print(' Parameters loaded.')
except yaml.YAMLError as exc:
print(exc)
m = Darknet(cfgfile)
region_loss = m.loss
m.load_weights(weightfile)
print('--- bn weight ---')
print(m.models[0][1].weight)
print('--- bn bias ---')
print(m.models[0][1].bias)
print('--- bn running_mean ---')
print(m.models[0][1].running_mean)
print('--- bn running_var ---')
print(m.models[0][1].running_var)
m.train()
if torch.cuda.is_available():
m = m.cuda()
optimizer = optim.SGD(m.parameters(), lr=1e-2, momentum=0.9, weight_decay=0.1)
img = Image.open(imgpath)
img = image2torch(img)
if torch.cuda.is_available():
img = img.cuda()
img = Variable(img)
target = Variable(label)
print('----- img ---------------------')
print(img.data.storage()[0:100])
print('----- target -----------------')
print(target.data.storage()[0:100])
class ObjectDetection:
def __init__(self, id):
# self.cap = cv2.VideoCapture(id)
self.cap = WebcamVideoStream(src = id).start()
self.cfgfile = "cfg/yolov3.cfg"
# self.cfgfile = 'cfg/yolov3-tiny.cfg'
self.weightsfile = "yolov3.weights"
# self.weightsfile = 'yolov3-tiny.weights'
self.confidence = float(0.5)
self.nms_thesh = float(0.4)
self.num_classes = 80
self.classes = load_classes('data/coco.names')
self.colors = pkl.load(open("pallete", "rb"))
self.model = Darknet(self.cfgfile)
self.CUDA = torch.cuda.is_available()
self.model.load_weights(self.weightsfile)
self.model.net_info["height"] = 160
self.inp_dim = int(self.model.net_info["height"])
self.width = 640 #640#
self.height = 480 #360#
print("Loading network.....")
if self.CUDA:
self.model.cuda()
print("Network successfully loaded")
assert self.inp_dim % 32 == 0
assert self.inp_dim > 32
self.model.eval()
def main(self):
q = queue.Queue()
def frame_render(queue_from_cam):
frame = self.cap.read()
frame = cv2.resize(frame,(self.width, self.height))
queue_from_cam.put(frame)
cam = threading.Thread(target=frame_render, args=(q,))
cam.start()
cam.join()
frame = q.get()
q.task_done()
fps = FPS().start()
try:
img, orig_im, dim = prep_image(frame, self.inp_dim)
im_dim = torch.FloatTensor(dim).repeat(1,2)
if self.CUDA: #### If you have a gpu properly installed then it will run on the gpu
im_dim = im_dim.cuda()
img = img.cuda()
# with torch.no_grad(): #### Set the model in the evaluation mode
output = self.model(Variable(img), self.CUDA)
output = write_results(output, self.confidence, self.num_classes, nms = True, nms_conf = self.nms_thesh) #### Localize the objects in a frame
output = output.type(torch.half)
if list(output.size()) == [1,86]:
pass
else:
output[:,1:5] = torch.clamp(output[:,1:5], 0.0, float(self.inp_dim))/self.inp_dim
# im_dim = im_dim.repeat(output.size(0), 1)
output[:,[1,3]] *= frame.shape[1]
output[:,[2,4]] *= frame.shape[0]
list(map(lambda x: write(x, frame, self.classes, self.colors),output))
x,y,w,h = b_boxes["bbox"][0],b_boxes["bbox"][1], b_boxes["bbox"][2], b_boxes["bbox"][3]
distance = (2 * 3.14 * 180) / (w + h * 360) * 1000 + 3 ### Distance measuring in Inch
feedback = ("{}".format(labels["Current Object"])+ " " +"is"+" at {} ".format(round(distance))+"Inches")
# speak.Speak(feedback) # If you are running this on linux based OS kindly use espeak. Using this speaking library in winodws will add unnecessary latency
# print(feedback)
except:
pass
fps.update()
fps.stop()
print("[INFO] elasped time: {:.2f}".format(fps.elapsed()))
print("[INFO] approx. FPS: {:.1f}".format(fps.fps()))
frame = cv2.putText(frame, str("{:.2f} Inches".format(distance)), (x,y), cv2.FONT_HERSHEY_DUPLEX, 0.6, (0,0,255), 1, cv2.LINE_AA)
ret, jpeg = cv2.imencode('.jpg', frame)
return jpeg.tostring()
def main():
args = arg_parse()
confidence = args.confidence
nms_thresh = args.nms_thresh
start = 0
CUDA = torch.cuda.is_available()
classes = load_classes("data/coco.names")
num_classes = len(classes)
# Set up the neural network
print("Loading network.....")
model = Darknet(args.cfgfile)
model.load_weights(args.weightsfile)
print("Network successfully loaded")
model.net_info["height"] = args.reso
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
# if there's a GPU available, put the model on GPU
if CUDA:
model.cuda()
# set the model in evaluation mode
model.eval()
def write(x, img, color):
c1 = tuple(x[1:3].int())
c2 = tuple(x[3:5].int())
cls = int(x[-1])
label = "{0}".format(classes[cls])
cv2.rectangle(img, c1, c2, color, 4)
t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 1, 1)[0]
c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4
cv2.rectangle(img, c1, c2, color, -1)
cv2.putText(img,
label, (c1[0], c1[1] + t_size[1] + 4),
cv2.FONT_HERSHEY_SIMPLEX,
0.5, (150, 150, 150),
thickness=1)
# detection phaase
cap = cv2.VideoCapture(0)
assert cap.isOpened(), "Cannot capture source"
frames = 0
start = time.time()
hsv_tuples = [(x / num_classes, 1., 1.) for x in range(num_classes)]
colors = list(map(lambda x: colorsys.hsv_to_rgb(*x), hsv_tuples))
colors = list(
map(lambda x: (int(x[0] * 200), int(x[1] * 200), int(x[2] * 200)),
colors))
np.random.seed(10000)
np.random.shuffle(colors)
np.random.seed(None) # reset seed to default.
while cap.isOpened():
ret, frame = cap.read()
if ret:
frame = cv2.resize(frame, dsize=(1280, 960))
img = prep_image(frame, inp_dim)
print(f"IMG_SHAPE: {img.shape}")
im_dim = frame.shape[1], frame.shape[0]
im_dim = torch.FloatTensor(im_dim).repeat(1, 2)
if CUDA:
im_dim = im_dim.cuda()
img = img.cuda()
with torch.no_grad():
outputs = model(Variable(img, volatile=True), CUDA)
outputs = write_results(outputs,
confidence,
num_classes,
nms_conf=nms_thresh)
if outputs != None:
im_dim = im_dim.repeat(outputs.size(0), 1)
scaling_factor = torch.min(inp_dim / im_dim, 1)[0].view(-1, 1)
outputs[:, [1, 3]] -= (
inp_dim - scaling_factor * im_dim[:, 0].view(-1, 1)) / 2
outputs[:, [2, 4]] -= (
inp_dim - scaling_factor * im_dim[:, 1].view(-1, 1)) / 2
outputs[:, 1:5] /= scaling_factor
for i in range(outputs.shape[0]):
outputs[i, [1, 3]] = torch.clamp(outputs[i, [1, 3]], 0.0,
im_dim[i, 0])
outputs[i, [2, 4]] = torch.clamp(outputs[i, [2, 4]], 0.0,
im_dim[i, 1])
for output in outputs:
color = colors[int(output[-1])]
write(output, frame, color)
cv2.imshow("frame", frame)
key = cv2.waitKey(1)
if key & 0xFF == ord('q'):
break
frames += 1
print(time.time() - start)
print("FPS of the video is {:5.2f}".format(frames /
(time.time() - start)))
else:
break
CUDA = torch.cuda.is_available()
num_classes = 80
bbox_attrs = 5 + num_classes
print("Loading network.....")
model = Darknet(args.cfgfile)
model.load_weights(args.weightsfile)
print("Network successfully loaded")
model.net_info["height"] = args.reso
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
if CUDA:
model.cuda().half()
# model(get_test_input(inp_dim, CUDA), CUDA)
# model.eval()
videofile = 'sample.mp4'
cap = cv2.VideoCapture(videofile)
assert cap.isOpened(), 'Cannot capture source'
frames = 0
start = time.time()
while cap.isOpened():
class ObjectDetection:
def __init__(self, id):
# self.cap = cv2.VideoCapture(id)
self.cap = WebcamVideoStream(src = id).start()
self.cfgfile = "cfg/yolov3.cfg"
# self.cfgfile = 'cfg/yolov3-tiny.cfg'
self.weightsfile = "yolov3.weights"
# self.weightsfile = 'yolov3-tiny.weights'
self.confidence = float(0.6)
self.nms_thesh = float(0.8)
self.num_classes = 80
self.classes = load_classes('data/coco.names')
self.colors = pkl.load(open("pallete", "rb"))
self.model = Darknet(self.cfgfile)
self.CUDA = torch.cuda.is_available()
self.model.load_weights(self.weightsfile)
self.model.net_info["height"] = 160
self.inp_dim = int(self.model.net_info["height"])
self.width = 1280 #640#1280
self.height = 720 #360#720
print("Loading network.....")
if self.CUDA:
self.model.cuda()
print("Network successfully loaded")
assert self.inp_dim % 32 == 0
assert self.inp_dim > 32
self.model.eval()
def main(self):
q = queue.Queue()
while True:
def frame_render(queue_from_cam):
frame = self.cap.read() # If you capture stream using opencv (cv2.VideoCapture()) the use the following line
# ret, frame = self.cap.read()
frame = cv2.resize(frame,(self.width, self.height))
queue_from_cam.put(frame)
cam = threading.Thread(target=frame_render, args=(q,))
cam.start()
cam.join()
frame = q.get()
q.task_done()
fps = FPS().start()
try:
img, orig_im, dim = prep_image(frame, self.inp_dim)
im_dim = torch.FloatTensor(dim).repeat(1,2)
if self.CUDA: #### If you have a gpu properly installed then it will run on the gpu
im_dim = im_dim.cuda()
img = img.cuda()
# with torch.no_grad(): #### Set the model in the evaluation mode
output = self.model(Variable(img), self.CUDA)
output = write_results(output, self.confidence, self.num_classes, nms = True, nms_conf = self.nms_thesh) #### Localize the objects in a frame
output = output.type(torch.half)
if list(output.size()) == [1,86]:
print(output.size())
pass
else:
output[:,1:5] = torch.clamp(output[:,1:5], 0.0, float(self.inp_dim))/self.inp_dim
# im_dim = im_dim.repeat(output.size(0), 1)
output[:,[1,3]] *= frame.shape[1]
output[:,[2,4]] *= frame.shape[0]
list(map(lambda boxes: write(boxes, frame, self.classes, self.colors),output))
except:
pass
fps.update()
fps.stop()
ret, jpeg = cv2.imencode('.jpg', frame)
print("[INFO] elasped time: {:.2f}".format(fps.elapsed()))
print("[INFO] approx. FPS: {:.1f}".format(fps.fps()))
return jpeg.tostring()
def predict():
target = os.path.join(APP_ROOT, 'static/')
print(target)
if not os.path.isdir(target):
os.mkdir(target)
else:
print("Couldn't create upload directory: {}".format(target))
print(request.files.getlist("file"))
for upload in request.files.getlist("file"):
print(upload)
print("{} is the file name".format(upload.filename))
filename = upload.filename
destination = "/".join([target, filename])
print("Accept incoming file:", filename)
print("Save it to:", destination)
upload.save(destination)
scales = "1,2,3"
print(filename)
images = "static/" + str(filename)
batch_size = int(1)
confidence = float(0.5)
nms_thesh = float(0.4)
start = 0
CUDA = torch.cuda.is_available()
num_classes = 80
classes = load_classes('data/coco.names')
print("Loading network.....")
model = Darknet("cfg/yolov3.cfg")
model.load_weights("yolov3.weights")
print("Network successfully loaded")
model.net_info["height"] = "416"
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
if CUDA:
model.cuda()
model.eval()
read_dir = time.time()
try:
imlist = [
osp.join(osp.realpath('.'), images, img)
for img in os.listdir(images)
if os.path.splitext(img)[1] == '.png' or os.path.splitext(img)[1]
== '.jpeg' or os.path.splitext(img)[1] == '.jpg'
]
except NotADirectoryError:
imlist = []
imlist.append(osp.join(osp.realpath('.'), images))
except FileNotFoundError:
print("No file or directory with the name {}".format(images))
exit()
load_batch = time.time()
batches = list(
map(prep_image, imlist, [inp_dim for x in range(len(imlist))]))
im_batches = [x[0] for x in batches]
orig_ims = [x[1] for x in batches]
im_dim_list = [x[2] for x in batches]
im_dim_list = torch.FloatTensor(im_dim_list).repeat(1, 2)
if CUDA:
im_dim_list = im_dim_list.cuda()
leftover = 0
if (len(im_dim_list) % batch_size):
leftover = 1
if batch_size != 1:
num_batches = len(imlist) // batch_size + leftover
im_batches = [
torch.cat(
(im_batches[i * batch_size:min((i + 1) *
batch_size, len(im_batches))]))
for i in range(num_batches)
]
i = 0
write = False
model(get_test_input(inp_dim, CUDA), CUDA)
start_det_loop = time.time()
objs = {}
f = open("result.txt", "w+")
for batch in im_batches:
start = time.time()
if CUDA:
batch = batch.cuda()
with torch.no_grad():
prediction = model(Variable(batch), CUDA)
prediction = write_results(prediction,
confidence,
num_classes,
nms=True,
nms_conf=nms_thesh)
if type(prediction) == int:
i += 1
continue
end = time.time()
prediction[:, 0] += i * batch_size
if not write:
output = prediction
write = 1
else:
output = torch.cat((output, prediction))
for im_num, image in enumerate(
imlist[i * batch_size:min((i + 1) * batch_size, len(imlist))]):
im_id = i * batch_size + im_num
objs = [classes[int(x[-1])] for x in output if int(x[0]) == im_id]
print("{0:20s} predicted in {1:6.3f} seconds".format(
image.split("/")[-1], (end - start) / batch_size))
print("{0:20s} {1:s}".format("Objects Detected:", " ".join(objs)))
#f.write(listToString(objs))
obj6 = []
for i in objs:
if i not in obj6:
obj6.append(i)
f.write(listToString(obj6))
print(f.read())
print("----------------------------------------------------------")
#i += 1
if CUDA:
torch.cuda.synchronize()
try:
output
except NameError:
print("No detections were made")
exit()
im_dim_list = torch.index_select(im_dim_list, 0, output[:, 0].long())
scaling_factor = torch.min(inp_dim / im_dim_list, 1)[0].view(-1, 1)
output[:, [1, 3]] -= (inp_dim -
scaling_factor * im_dim_list[:, 0].view(-1, 1)) / 2
output[:, [2, 4]] -= (inp_dim -
scaling_factor * im_dim_list[:, 1].view(-1, 1)) / 2
output[:, 1:5] /= scaling_factor
for i in range(output.shape[0]):
output[i, [1, 3]] = torch.clamp(output[i, [1, 3]], 0.0, im_dim_list[i,
0])
output[i, [2, 4]] = torch.clamp(output[i, [2, 4]], 0.0, im_dim_list[i,
1])
output_recast = time.time()
class_load = time.time()
colors = pkl.load(open("pallete", "rb"))
draw = time.time()
def write(x, batches, results):
c1 = tuple(x[1:3].int())
c2 = tuple(x[3:5].int())
img = results[int(x[0])]
cls = int(x[-1])
label = "{0}".format(classes[cls])
color = random.choice(colors)
cv2.rectangle(img, c1, c2, color, 2)
t_size = cv2.getTextSize(label, cv2.FONT_HERSHEY_PLAIN, 1, 1)[0]
c2 = c1[0] + t_size[0] + 3, c1[1] + t_size[1] + 4
cv2.rectangle(img, c1, c2, color, -1)
cv2.putText(img, label, (c1[0], c1[1] + t_size[1] + 4),
cv2.FONT_HERSHEY_PLAIN, 1, [225, 255, 255], 1)
return img
list(map(lambda x: write(x, im_batches, orig_ims), output))
det_names = pd.Series(imlist).apply(
lambda x: "{}/{}".format("static",
x.split("/")[-1]))
list(map(cv2.imwrite, det_names, orig_ims))
end = time.time()
torch.cuda.empty_cache()
main1()
main2()
with open("result.txt", "r") as read_file:
with open("result2.txt", "w") as write_file:
write_file.write(read_file.read().replace(" ", '\n'))
with open("hashesfromcaption.txt", "r") as read_file:
with open("hsh.txt", "w") as write_file:
write_file.write(read_file.read().replace(" ", '_'))
with open("hsh.txt", "r") as read_file:
with open("h.txt", "w") as write_file:
write_file.write(read_file.read().replace("#", '\n#'))
text = open('output.txt', 'r+')
content = text.read()
text.close()
objk = open('result.txt', 'r+')
contentobjk = objk.read()
objk.close()
hashcode = open('h.txt', 'r+')
h = hashcode.read()
hashcode.close()
return render_template("results.html",
image_name=filename,
text=content,
objects=contentobjk,
last=h)
CUDA = torch.cuda.is_available()
bbox_attrs = 5 + num_classes
print("Loading network.....")
model = Darknet(args.cfgfile)
model.load_weights(args.weightsfile)
print("Network successfully loaded")
model.net_info["height"] = args.reso
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
if CUDA:
model.cuda()
model(get_test_input(inp_dim, CUDA), CUDA)
model.eval()
videofile = args.video
cap = cv2.VideoCapture(videofile)
video_frame_cnt = int(cap.get(7))
video_width = int(cap.get(3))
video_height = int(cap.get(4))
video_fps = int(cap.get(5))
fourcc = cv2.VideoWriter_fourcc('m', 'p', '4', 'v')
videoWriter = cv2.VideoWriter('video_result.mp4', fourcc, video_fps, (video_width, video_height))
class Detector(object):
def __init__(self, model_def_file='', weights_file=''):
self.model_def_file = model_def_file
self.weights_file = weights_file
self.model = Darknet(self.model_def_file)
self.model.load_weights(self.weights_file)
self.CUDA = True
if self.CUDA:
self.model.cuda()
print('load network finish')
self.confidence = 0.5
self.nms_thresh = 0.4
self.num_classes = 80
self.yolo_dir = '/home/yfji/SourceCode/pytorch-yolo-v3'
self.classes = util.load_classes(op.join('data/coco.names'))
self.colors = pickle.load(open(op.join(self.yolo_dir, 'pallete',
'rb')))
def detect(self, image):
prediction = self.model(Variable(image), self.CUDA)
output = self.filter_results(prediction) #list of [score, x1,y1,x2,y2]
return output
def filter_results(self, prediction, nms=True):
conf_mask = (prediction[:, :, 4] >
self.confidence).float().unsqueeze(2)
prediction = prediction * conf_mask
try:
torch.nonzero(prediction[:, :, 4]).transpose(0, 1).contiguous()
except:
return 0
box_a = prediction.new(prediction.shape)
box_a[:, :, 0] = (prediction[:, :, 0] - prediction[:, :, 2] / 2)
box_a[:, :, 1] = (prediction[:, :, 1] - prediction[:, :, 3] / 2)
box_a[:, :, 2] = (prediction[:, :, 0] + prediction[:, :, 2] / 2)
box_a[:, :, 3] = (prediction[:, :, 1] + prediction[:, :, 3] / 2)
prediction[:, :, :4] = box_a[:, :, :4]
batch_size = prediction.size(0)
output = prediction.new(1, prediction.size(2) + 1)
write = False
for ind in range(batch_size):
image_pred = prediction[ind]
max_conf, max_conf_score = torch.max(
image_pred[:, 5:5 + self.num_classes], 1)
max_conf = max_conf.float().unsqueeze(1)
max_conf_score = max_conf_score.float().unsqueeze(1)
seq = (image_pred[:, :5], max_conf, max_conf_score)
image_pred = torch.cat(seq, 1)
non_zero_ind = (torch.nonzero(image_pred[:, 4]))
try:
image_pred_ = image_pred[non_zero_ind.squeeze(), :].view(-1, 7)
except:
continue
img_classes = util.unique(image_pred_[:, -1])
for cls in img_classes:
#get the detections with one particular class
cls_mask = image_pred_ * (image_pred_[:, -1]
== cls).float().unsqueeze(1)
class_mask_ind = torch.nonzero(cls_mask[:, -2]).squeeze()
image_pred_class = image_pred_[class_mask_ind].view(-1, 7)
conf_sort_index = torch.sort(image_pred_class[:, 4],
descending=True)[1]
image_pred_class = image_pred_class[conf_sort_index]
idx = image_pred_class.size(0)
if nms:
for i in range(idx):
try:
ious = util.bbox_iou(
image_pred_class[i].unsqueeze(0),
image_pred_class[i + 1:])
except ValueError:
break
except IndexError:
break
iou_mask = (ious <
self.nms_thresh).float().unsqueeze(1)
image_pred_class[i + 1:] *= iou_mask
non_zero_ind = torch.nonzero(
image_pred_class[:, 4]).squeeze()
image_pred_class = image_pred_class[non_zero_ind].view(
-1, 7)
batch_ind = image_pred_class.new(image_pred_class.size(0),
1).fill_(ind)
seq = batch_ind, image_pred_class
if not write:
output = torch.cat(seq, 1)
write = True
else:
out = torch.cat(seq, 1)
output = torch.cat((output, out))
return output
def start(self):
# Inicializacion de variables globales
global classes, BBox, colors, phase, frame, initBBox, true_class_filter
# PREPARACION DE LA FASE DE DETECCION
CUDA = torch.cuda.is_available()
text = 'No class filter selected'
classes = load_classes('model/{}/model.names'.format(
self.model_folder))
colors = pkl.load(open('pallete', 'rb'))
num_classes = len(classes)
if [i for i in self.class_filter if not (i in classes)]:
if self.label_info:
text = 'WARNING: {} class/classes are not included in the selected model. Updating the searching list...'.format(
[i for i in self.class_filter if not (i in classes)])
self.label_info.setText(text)
else:
print(
'WARNING: {} class/classes are not included in the selected model. Updating the searching list...'
.format(
[i for i in self.class_filter if not (i in classes)]))
true_class_filter = [i for i in self.class_filter if (i in classes)]
# Configuracion de la red
if self.label_info:
text += '\nLoading network...'
self.label_info.setText(text)
else:
print('Loading network.....')
model = Darknet('model/{}/model.cfg'.format(self.model_folder))
model.load_weights('model/{}/model.weights'.format(self.model_folder))
if self.label_info:
text += '\nNetwork succesfully loaded'
self.label_info.setText(text)
else:
print('Network successfully loaded')
model.net_info['height'] = self.reso_det
inp_dim_det = int(model.net_info['height'])
assert inp_dim_det % 32 == 0
assert inp_dim_det > 32
# Si hay un dispositivo CUDA se carga en el el modelo
if CUDA:
model.cuda()
# Modelo en modo de evaluacion
model.eval()
# PREPARACION DE LA FASE DE TRACKING
inp_dim_track = int(self.reso_track)
OPENCV_OBJECT_TRACKERS = {
'csrt': cv2.TrackerCSRT_create,
'kcf': cv2.TrackerKCF_create,
'boosting': cv2.TrackerBoosting_create,
'mil': cv2.TrackerMIL_create,
'tld': cv2.TrackerTLD_create,
'medianflow': cv2.TrackerMedianFlow_create,
'mosse': cv2.TrackerMOSSE_create
}
# INICIALIZACION DE LA FUENTE
if self.source == '0' or self.source == '1':
self.cap = cv2.VideoCapture(int(self.source))
mode = 'cam'
self.window_name = 'Camera ' + self.source
else:
if self.label_info: # via GUI se obtiene el path completo
self.cap = cv2.VideoCapture(self.source)
else: # via terminal solo escribimos el nombre del archivo
self.cap = cv2.VideoCapture('videos/{}'.format(self.source))
mode = 'file'
self.window_name = self.source
assert self.cap.isOpened(), 'Cannot capture source'
phase = 'det'
initBBox = []
cont = 0
frames = 0
cv2.namedWindow(self.window_name)
cv2.setMouseCallback(self.window_name, click_det2track)
while self.cap.isOpened():
grab, frame = self.cap.read()
start = time.time()
if grab:
# Fase de deteccion
if phase == 'det':
if mode == 'cam':
img = prep_image_c(frame, inp_dim_det)
elif mode == 'file':
img = prep_image_f(frame, inp_dim_det)
im_dim = frame.shape[1], frame.shape[0]
im_dim = torch.FloatTensor(im_dim).repeat(1, 2)
if CUDA:
im_dim = im_dim.cuda()
img = img.cuda()
# Inicializacion la lista de BBox detectadas
BBox = []
output = model.forward(Variable(img), CUDA)
output = write_results(output,
self.confidence,
num_classes,
nms_conf=self.nms_thresh)
if type(output) == int:
frames += 1
cv2.imshow(self.window_name, frame)
key = cv2.waitKey(1)
if key & 0xFF == ord('q'):
break
continue
if mode == 'cam':
output[:, 1:5] = torch.clamp(output[:, 1:5], 0.0,
float(inp_dim_det))
im_dim = im_dim.repeat(output.size(0), 1) / inp_dim_det
output[:, 1:5] *= im_dim
elif mode == 'file':
im_dim = im_dim.repeat(output.size(0), 1)
scaling_factor = torch.min(inp_dim_det / im_dim,
1)[0].view(-1, 1)
output[:, [1, 3]] -= (inp_dim_det - scaling_factor *
im_dim[:, 0].view(-1, 1)) / 2
output[:, [2, 4]] -= (inp_dim_det - scaling_factor *
im_dim[:, 1].view(-1, 1)) / 2
output[:, 1:5] /= scaling_factor
for i in range(output.shape[0]):
output[i, [1, 3]] = torch.clamp(
output[i, [1, 3]], 0.0, im_dim[i, 0])
output[i, [2, 4]] = torch.clamp(
output[i, [2, 4]], 0.0, im_dim[i, 1])
list(map(lambda x: write(x, frame), output))
cv2.imshow(self.window_name, frame)
key = cv2.waitKey(1)
if key & 0xFF == ord('q'):
break
frames += 1
if self.label_info:
self.label_info.setText(
text + '\nDETECTION PHASE:' +
'\n {0: .2f} fps'.format(
float(1 / (time.time() - start))))
# Fase de tracking
elif phase == 'track':
ratio = frame.shape[0] / inp_dim_track
img = imutils.resize(frame, height=inp_dim_track)
if initBBox:
(success, box) = tracker.update(img)
if success:
cont = 0
(x, y, w, h) = [int(v) for v in box]
x, y, w, h = prep_rect(x, y, w, h, ratio)
cv2.rectangle(frame, (x, y), (x + w, y + h),
(0, 255, 0), 2)
else:
cont += 1
if self.label_info:
self.label_info.setText(
text + '\nTRACKING PHASE' +
'\nObject lost ({})'.format(cont))
else:
print('Object lost ', cont)
else:
(x, y, w, h) = [int(v) for v in track_rect]
initBBox = (prep_rect(x, y, w, h, float(1 / ratio)))
tracker = OPENCV_OBJECT_TRACKERS[self.tracker_alg]()
tracker.init(img, initBBox)
if cont > 100:
phase = 'det'
cont = 0
initBBox = []
cv2.imshow(self.window_name, frame)
key = cv2.waitKey(1)
if key & 0xFF == ord('q'):
break
frames += 1
if self.label_info:
self.label_info.setText(
text + '\nTRACKING PHASE:' +
'\n {0: .2f} fps'.format(
float(1 / (time.time() - start))))
else:
break
else:
break
if not self.label_info:
cv2.destroyWindow(self.window_name)
self.cap.release()
torch.cuda.empty_cache()
class ssp_rosbag:
def __init__(self):
rospy.init_node('eval_baseline', anonymous=True)
##############################################################################
##############################################################################
##############################################################################
self.b_first_rb_loop = True
self.first_time = None
self.ns = rospy.get_param('~ns') # robot namespace
modelcfg = rospy.get_param('~modelcfg')
weightfile = rospy.get_param('~weightfile')
datacfg = rospy.get_param('~datacfg')
rb_name = rospy.get_param('~rb_name')
self.ado_names = [rospy.get_param('~tracked_name')]
# Parse configuration files
data_options = read_data_cfg(datacfg)
valid_images = data_options['valid']
if 'mesh' in data_options:
meshname = data_options['mesh']
else:
meshname = None
assert ('box_length' in data_options)
box_length = float(data_options['box_length'])
box_width = float(data_options['box_width'])
box_height = float(data_options['box_height'])
self.ego_name = data_options['name']
gpus = data_options['gpus']
self.im_width = int(data_options['width'])
self.im_height = int(data_options['height'])
# Parameters
seed = int(time.time())
os.environ['CUDA_VISIBLE_DEVICES'] = gpus
torch.cuda.manual_seed(seed)
self.num_classes = 1
# Read object model information, get 3D bounding box corners
if meshname is None:
# vertices must be 4 x N for compute_projections to work later
vertices = np.array(
[[box_length / 2, box_width / 2, box_height / 2, 1.],
[box_length / 2, box_width / 2, -box_height / 2, 1.],
[box_length / 2, -box_width / 2, -box_height / 2, 1.],
[box_length / 2, -box_width / 2, box_height / 2, 1.],
[-box_length / 2, -box_width / 2, box_height / 2, 1.],
[-box_length / 2, -box_width / 2, -box_height / 2, 1.],
[-box_length / 2, box_width / 2, -box_height / 2, 1.],
[-box_length / 2, box_width / 2, box_height / 2, 1.]]).T
self.diam = float(data_options['diam'])
else:
mesh = MeshPly(meshname)
vertices = np.c_[np.array(mesh.vertices),
np.ones((len(mesh.vertices), 1))].transpose()
try:
self.diam = float(data_options['diam'])
except:
self.diam = calc_pts_diameter(np.array(mesh.vertices))
self.vertices = vertices
self.corners3D = get_3D_corners(vertices)
# Specify model, load pretrained weights, pass to GPU and set the module in evaluation mode
torch.set_grad_enabled(False) # since we are just doing forward passes
self.model = Darknet(modelcfg)
self.model.load_weights(weightfile)
self.model.cuda()
self.model.eval()
self.shape = (self.model.test_width, self.model.test_height)
num_keypoints = self.model.num_keypoints
num_labels = num_keypoints * 2 + 3 # +2 for width, height, +1 for class label
##############################################################################
##############################################################################
##############################################################################
self.result_list = [] # save the results as they are processed
self.itr = 0
self.time_prev = -1
self.bridge = CvBridge()
self.pose_buffer_len = 20
self.ado_pose_msg_buf = []
self.ego_pose_msg_buf = []
self.ego_pose_est_msg_buf = []
self.ego_pose_est_time_msg_buf = []
self.ado_pose_time_msg_buf = []
self.ego_pose_time_msg_buf = []
# Create camera (camera extrinsics from quad7.param in the msl_raptor project):
self.tf_cam_ego = np.eye(4)
self.tf_cam_ego[0:3,
3] = np.asarray([0.01504337, -0.06380886, -0.13854437])
self.tf_cam_ego[0:3, 0:3] = np.reshape([
-6.82621737e-04, -9.99890488e-01, -1.47832690e-02, 3.50423970e-02,
1.47502748e-02, -9.99276969e-01, 9.99385593e-01, -1.20016936e-03,
3.50284906e-02
], (3, 3))
# Correct Rotation w/ Manual Calibration
Angle_x = 8. / 180.
Angle_y = 8. / 180.
Angle_z = 0. / 180.
R_deltax = np.array([[1., 0., 0.],
[0., np.cos(Angle_x), -np.sin(Angle_x)],
[0., np.sin(Angle_x),
np.cos(Angle_x)]])
R_deltay = np.array([[np.cos(Angle_y), 0.,
np.sin(Angle_y)], [0., 1., 0],
[-np.sin(Angle_y), 0.,
np.cos(Angle_y)]])
R_deltaz = np.array([[np.cos(Angle_z), -np.sin(Angle_z), 0.],
[np.sin(Angle_z),
np.cos(Angle_z), 0.], [0., 0., 1.]])
R_delta = np.dot(R_deltax, np.dot(R_deltay, R_deltaz))
self.tf_cam_ego[0:3, 0:3] = np.matmul(R_delta, self.tf_cam_ego[0:3,
0:3])
#########################################################################################
camera_info = rospy.wait_for_message(self.ns + '/camera/camera_info',
CameraInfo, 30)
self.K = np.reshape(camera_info.K, (3, 3))
self.dist_coefs = np.reshape(camera_info.D, (5, ))
self.new_camera_matrix, _ = cv2.getOptimalNewCameraMatrix(
self.K, self.dist_coefs, (camera_info.width, camera_info.height),
0, (camera_info.width, camera_info.height))
self.log_out_dir = '/mounted_folder/ssp_logs'
# ssp_log_name = self.log_out_dir + "/log_" + rb_name.split("_")[-1] + "_SSP.log"
# param_log_name = self.log_out_dir + "/log_" + rb_name.split("_")[-1] + "_PARAM.log"
# self.logger = raptor_logger(source="SSP", mode="write", ssp_fn=ssp_log_name, param_fn=param_log_name)
base_path = self.log_out_dir + "/log_" + rb_name.split("_")[-1]
self.rb_name = rb_name
self.bb_3d_dict_all = {
self.ado_names[0]: [box_length, box_width, box_height, self.diam]
}
self.logger = RaptorLogger(mode="write",
names=self.ado_names,
base_path=base_path,
b_ssp=True)
# Write params to log file ########################################################################################################
param_data = {}
if self.new_camera_matrix is not None:
param_data['K'] = np.array([
self.new_camera_matrix[0, 0], self.new_camera_matrix[1, 1],
self.new_camera_matrix[0, 2], self.new_camera_matrix[1, 2]
])
else:
param_data['K'] = np.array(
[self.K[0, 0], self.K[1, 1], self.K[0, 2], self.K[1, 2]])
param_data['3d_bb_dims'] = np.array(
[box_length, box_width, box_height, self.diam])
param_data['tf_cam_ego'] = np.reshape(copy(self.tf_cam_ego), (16, ))
# self.logger.write_data_to_log(log_data, mode='prms')
self.logger.write_params(param_data)
###################################################################################################################################
self.t0 = None
self.time_arr = []
# self.raptor_metrics = pose_metric_tracker(px_thresh=5, prct_thresh=10, trans_thresh=0.05, ang_thresh=5, name=self.name, diam=self.diam)
self.raptor_metrics = PoseMetricTracker(px_thresh=5,
prct_thresh=10,
trans_thresh=0.05,
ang_thresh=5,
names=self.ado_names,
bb_3d_dict=self.bb_3d_dict_all)
rospy.Subscriber(self.ns + '/mavros/vision_pose/pose',
PoseStamped,
self.ego_pose_gt_cb,
queue_size=10) # optitrack pose
rospy.Subscriber(self.ns + '/mavros/local_position/pose',
PoseStamped,
self.ego_pose_est_cb,
queue_size=10) # onboard ekf pose est
rospy.Subscriber('/quad4' + '/mavros/vision_pose/pose',
PoseStamped,
self.ado_pose_gt_cb,
queue_size=10) # optitrack pose
rospy.Subscriber(self.ns + '/camera/image_raw',
ROS_IMAGE,
self.image_cb,
queue_size=1,
buff_size=2**21)
def ado_pose_gt_cb(self, msg):
# if self.first_time is not None and self.first_time >= msg.header.stamp.to_sec():
# return
self.ado_pose_gt_rosmsg = msg.pose
pose_tm = msg.header.stamp.to_sec()
self.ado_pose_msg_buf.append(msg)
self.ado_pose_time_msg_buf.append(pose_tm)
def ego_pose_gt_cb(self, msg):
# if self.first_time is not None and self.first_time >= msg.header.stamp.to_sec():
# return
self.ego_pose_gt_rosmsg = msg.pose
pose_tm = msg.header.stamp.to_sec()
self.ego_pose_msg_buf.append(msg)
self.ego_pose_time_msg_buf.append(pose_tm)
def ego_pose_est_cb(self, msg):
# if self.first_time is not None and self.first_time >= msg.header.stamp.to_sec():
# return
self.ego_pose_est_rosmsg = msg.pose
pose_tm = msg.header.stamp.to_sec()
self.ego_pose_est_msg_buf.append(msg)
self.ego_pose_est_time_msg_buf.append(pose_tm)
def image_cb(self, msg):
"""
Maintains a buffer of images & times. The first element is the earliest.
Stored in a way to interface with a quick method for finding closest match by time.
"""
tic = time.time()
img_tm = msg.header.stamp.to_sec()
if len(program.result_list) > 0 and img_tm <= self.result_list[-1][5]:
return
if self.t0 is None:
self.t0 = img_tm
img_cv2 = self.bridge.imgmsg_to_cv2(msg,
desired_encoding="passthrough")
img_cv2 = cv2.undistort(img_cv2, self.K, self.dist_coefs, None,
self.new_camera_matrix)
img_pil = PIL.Image.fromarray(img_cv2).resize(self.shape)
img = Variable(transforms.ToTensor()(img_pil).resize(
1, 3, img_pil.size[0], img_pil.size[1]).cuda(),
volatile=True)
with torch.no_grad():
output = self.model(img).data # Forward pass
# Using confidence threshold, eliminate low-confidence predictions
box_pr = get_region_boxes(output, self.num_classes,
self.model.num_keypoints)
# Denormalize the corner predictions
corners2D_pr = np.array(np.reshape(box_pr[:18], [-1, 2]),
dtype='float32')
corners2D_pr[:, 0] = corners2D_pr[:, 0] * self.im_width
corners2D_pr[:, 1] = corners2D_pr[:, 1] * self.im_height
# Compute [R|t] by pnp
R_pr, t_pr = pnp(
np.array(np.transpose(
np.concatenate((np.zeros((3, 1)), self.corners3D[:3, :]),
axis=1)),
dtype='float32'), corners2D_pr,
np.array(self.K, dtype='float32'))
tf_cam_ado_est = rotm_and_t_to_tf(R_pr, t_pr)
if len(self.ado_pose_time_msg_buf) == 0 or len(
self.ego_pose_time_msg_buf) == 0 or len(
self.ego_pose_est_time_msg_buf) == 0:
print("still waiting for other rosbag messages")
return
ado_msg, _ = find_closest_by_time(img_tm,
self.ado_pose_time_msg_buf,
message_list=self.ado_pose_msg_buf)
ego_gt_msg, _ = find_closest_by_time(
img_tm,
self.ego_pose_time_msg_buf,
message_list=self.ego_pose_msg_buf)
ego_est_msg, _ = find_closest_by_time(
img_tm,
self.ego_pose_est_time_msg_buf,
message_list=self.ego_pose_est_msg_buf)
tf_w_ado_gt = pose_to_tf(ado_msg.pose)
tf_w_ego_gt = pose_to_tf(ego_gt_msg.pose)
tf_w_ego_est = pose_to_tf(ego_est_msg.pose)
tf_w_cam_gt = tf_w_ego_gt @ invert_tf(self.tf_cam_ego)
tf_w_ado_est = tf_w_cam_gt @ tf_cam_ado_est
quat_pr = rotm_to_quat(tf_w_ado_est[0:3, 0:3])
state_pr = np.concatenate((tf_w_ado_est[0:3,
3], quat_pr)) # shape = (7,)
b_remove_yaw = True
if b_remove_yaw:
quat_pr_with_yaw = quat_pr # quat with yaw
quat_gt = rotm_to_quat(tf_w_ado_gt[0:3, 0:3]) # quat with yaw
quat_pr = remove_yaw(quat_pr) # remove yaw
quat_gt = remove_yaw(quat_gt) # remove yaw
tf_w_ado_est[0:3, 0:3] = quat_to_rotm(quat_pr) # update tf
tf_w_ado_gt[0:3, 0:3] = quat_to_rotm(quat_gt) # update tf
img_to_save = copy(np.array(img.cpu()))
self.result_list.append(
(state_pr, copy(tf_w_ado_est), copy(tf_w_ado_gt),
copy(corners2D_pr), img_to_save, img_tm, time.time(), copy(R_pr),
copy(t_pr), invert_tf(tf_w_cam_gt), copy(tf_w_ego_gt),
copy(tf_w_ego_est)))
del img
self.itr += 1
self.time_arr.append(time.time() - tic)
if self.itr > 0 and self.itr % 50 == 0:
print("Finished processing image #{}, mean time: {}".format(
self.itr, np.mean(self.time_arr)))
torch.cuda.empty_cache()
def post_process_data(self):
print("Post-processing data now ({} itrs)".format(len(
self.result_list)))
b_save_bb_imgs = True
name = self.ado_names[0]
bb_im_path = os.path.dirname(os.path.relpath(
__file__)) + '/output_imgs' # PATH MUST BE RELATIVE
create_dir_if_missing(bb_im_path)
N = len(self.result_list)
# To save
trans_dist = 0.0
angle_dist = 0.0
pixel_dist = 0.0
testing_samples = 0.0
testing_error_trans = 0.0
testing_error_angle = 0.0
testing_error_pixel = 0.0
errs_2d = []
errs_3d = []
errs_trans = []
errs_angle = []
errs_corner2D = []
preds_trans = []
preds_rot = []
preds_corners2D = []
gts_trans = []
gts_rot = []
gts_corners2D = []
corners2D_gt = None
log_data = {}
for i, res in enumerate(self.result_list):
# extract / compute values for comparison
state_pr, tf_w_ado_est, tf_w_ado_gt, corners2D_pr, img, img_tm, sys_time, R_cam_ado_pr, t_cam_ado_pr, tf_cam_w_gt, tf_w_ego_gt, tf_w_ego_est = res
tf_cam_ado_gt = tf_cam_w_gt @ tf_w_ado_gt
R_cam_ado_gt = tf_cam_ado_gt[0:3, 0:3]
t_cam_ado_gt = tf_cam_ado_gt[0:3, 3].reshape(t_cam_ado_pr.shape)
if img_tm - self.t0 > 34 and self.rb_name == "rosbag_for_post_process_2019-12-18-02-10-28":
print("STOPPING EARLY")
break # quad crashes
Rt_cam_ado_gt = np.concatenate((R_cam_ado_gt, t_cam_ado_gt),
axis=1)
Rt_cam_ado_pr = np.concatenate((R_cam_ado_pr, t_cam_ado_pr),
axis=1)
corners2D_gt = compute_projection(
np.hstack((np.reshape([0, 0, 0, 1], (4, 1)), self.vertices)),
Rt_cam_ado_gt, self.new_camera_matrix).T
if b_save_bb_imgs:
draw_2d_proj_of_3D_bounding_box(img,
corners2D_pr,
corners2D_gt=corners2D_gt,
epoch=None,
batch_idx=None,
detect_num=i,
im_save_dir=bb_im_path)
if self.raptor_metrics is not None:
# self.raptor_metrics.update_all_metrics(vertices=self.vertices, R_gt=R_gt, t_gt=t_gt, R_pr=R_pr, t_pr=t_pr, K=self.new_camera_matrix)
self.raptor_metrics.update_all_metrics(
name=name,
vertices=self.vertices,
tf_w_cam=invert_tf(tf_cam_w_gt),
R_cam_ado_gt=R_cam_ado_gt,
t_cam_ado_gt=t_cam_ado_gt,
R_cam_ado_pr=R_cam_ado_pr,
t_cam_ado_pr=t_cam_ado_pr,
K=self.new_camera_matrix)
# Write data to log file #############################
log_data['time'] = img_tm - self.t0
log_data['state_est'] = tf_to_state_vec(tf_w_ado_est)
log_data['state_gt'] = tf_to_state_vec(tf_w_ado_gt)
log_data['ego_state_est'] = tf_to_state_vec(tf_w_ego_est)
log_data['ego_state_gt'] = tf_to_state_vec(tf_w_ego_gt)
corners3D_pr = (tf_w_ado_est @ self.vertices)[0:3, :]
corners3D_gt = (tf_w_ado_gt @ self.vertices)[0:3, :]
log_data['corners_3d_est'] = np.reshape(corners3D_pr,
(corners3D_pr.size, ))
log_data['corners_3d_gt'] = np.reshape(corners3D_gt,
(corners3D_gt.size, ))
log_data['proj_corners_est'] = np.reshape(
self.raptor_metrics.proj_2d_pr[name].T,
(self.raptor_metrics.proj_2d_pr[name].size, ))
log_data['proj_corners_gt'] = np.reshape(
self.raptor_metrics.proj_2d_gt[name].T,
(self.raptor_metrics.proj_2d_gt[name].size, ))
log_data['x_err'] = tf_w_ado_est[0, 3] - tf_w_ado_gt[0, 3]
log_data['y_err'] = tf_w_ado_est[1, 3] - tf_w_ado_gt[1, 3]
log_data['z_err'] = tf_w_ado_est[2, 3] - tf_w_ado_gt[2, 3]
log_data['ang_err'] = calcAngularDistance(tf_w_ado_est[0:3, 0:3],
tf_w_ado_gt[0:3, 0:3])
log_data['pix_err'] = np.mean(
la.norm(self.raptor_metrics.proj_2d_pr[name] -
self.raptor_metrics.proj_2d_gt[name],
axis=0))
log_data['add_err'] = np.mean(
la.norm(corners3D_pr - corners3D_gt, axis=0))
log_data['measurement_dist'] = la.norm(tf_w_ego_gt[0:3, 3] -
tf_w_ado_gt[0:3, 3])
self.logger.write_data_to_log(log_data, name, mode='ssp')
self.logger.write_data_to_log(log_data, name, mode='ssperr')
if np.any(np.isnan(corners3D_pr)) or np.any(
np.isnan(corners3D_gt)) or np.any(
np.isnan(self.raptor_metrics.proj_2d_pr[name])
): #or la.norm(tf_cam_ado_gt[0:3, 3] - t_cam_ado_pr) > 10:
print("ISSUE DETECTED!!")
pdb.set_trace()
######################################################
if self.raptor_metrics is not None:
self.raptor_metrics.calc_final_metrics()
self.raptor_metrics.print_final_metrics()
self.logger.close_files()
print("done with post process!")
def truths_length(self, truths, max_num_gt=50):
for i in range(max_num_gt):
if truths[i][1] == 0:
return i
def run(self):
rate = rospy.Rate(100)
b_flag = True
while not rospy.is_shutdown():
try:
rate.sleep()
except: # this will happen if the clock goes backwards (i.e. rosbag loops)
self.post_process_data()
return
test_loader = torch.utils.data.DataLoader(InriaDataset(img_dir,
lab_dir,
shuffle=True),
batch_size=3,
shuffle=True)
cfgfile = "cfg/yolov2.cfg"
weightfile = "weights/yolov2.weights"
printfile = "non_printability/30values.txt"
patch_size = 400
darknet_model = Darknet(cfgfile)
darknet_model.load_weights(weightfile)
darknet_model = darknet_model.cuda()
patch_applier = PatchApplier().cuda()
patch_transformer = PatchTransformer().cuda()
prob_extractor = MaxProbExtractor(0, 80).cuda()
nms_calculator = NMSCalculator(printfile, patch_size)
total_variation = TotalVariation()
'''
img = Image.open('data/horse.jpg').convert('RGB')
img = img.resize((darknet_model.width, darknet_model.height))
width = img.width
height = img.height
img = torch.ByteTensor(torch.ByteStorage.from_buffer(img.tobytes()))
img = img.view(height, width, 3).transpose(0, 1).transpose(0, 2).contiguous()
img = img.view(1, 3, height, width)
img = img.float().div(255.0)
img = torch.autograd.Variable(img)
#Set up the neural network
print("Loading network.....")
model = Darknet(args.cfgfile)
model.load_weights(args.weightsfile)
print("Network successfully loaded")
model.net_info["height"] = args.reso
inp_dim = int(model.net_info["height"])
assert inp_dim % 32 == 0
assert inp_dim > 32
#If there's a GPU availible, put the model on GPU
if CUDA:
model.cuda()
#Set the model in evaluation mode
model.eval()
read_dir = time.time()
#Detection phase
try:
imlist = [osp.join(osp.realpath('.'), images, img) for img in os.listdir(images)]
except NotADirectoryError:
imlist = []
imlist.append(osp.join(osp.realpath('.'), images))
except FileNotFoundError:
print ("No file or directory with the name {}".format(images))
exit()
