def train():
# Evaluation pipeline
files = glob.glob(
os.path.join('/home/iotsc_group1/ChangxingDENG/det/datasets/',
'PretrainImageNet', 'ILSVRC2012_img_val', '*.JPEG'))
files = sorted(files,
key=lambda f: f.split('/')[-1].split('_')[-1].split('.')[0])
labels = loadlabel(
os.path.join(
'/home/iotsc_group1/ChangxingDENG/det/datasets/',
'PretrainImageNet',
'ILSVRC2012_devkit_t12/data/ILSVRC2012_validation_ground_truth.txt'
))
eval_pipeline = EvalImageDecoderPipeline(files=files, labels=labels)
eval_pipeline.build()
eval_pii = pytorchIterator(eval_pipeline,
last_batch_policy=LastBatchPolicy.PARTIAL,
reader_name='Reader',
auto_reset=True)
model = Darknet()
state_dict = torch.load(
'logs/PretrainImageNet_20210316173822_1/13_70056.pth')
model.load_state_dict(state_dict=state_dict)
model = model.cuda()
criterion = nn.CrossEntropyLoss()
model.eval()
epoch_loss = 0
prediciton = []
target = []
with torch.no_grad():
for iter, data in enumerate(eval_pii):
x = data[0]['data']
label = data[0]['label'].squeeze(-1).long().cuda()
output = model(x)
loss = criterion(output, label).item()
epoch_loss += loss * x.shape[0]
prediciton.append(output)
target.append(label)
loss = epoch_loss / 50000
prediciton = torch.cat(prediciton, dim=0)
target = torch.cat(target, dim=0)
acc = top1accuracy(prediciton, target)
acctop5 = top5accuracy(prediciton, target)
print(f'Top1 ACC: {acc} Top5 ACC {acctop5} loss: {loss}')
def detect_images(widget=None):
if not widget:
args = arg_parse()
else:
args = widget.args
read_dir = time.time()
print_info(widget, False, "info", "Reading addresses.....")
images = args.images
im_list = []
img = None
try:
for img in images:
if os.path.isabs(img):
im_list.append(img)
else:
im_list.append(osp.join(osp.realpath('.'), img))
except FileNotFoundError:
print_info(widget, True, "error", "No file or directory with the name {}".format(img))
if not os.path.exists(args.det):
os.makedirs(args.det)
print_info(widget, False, "info", "Finished reading addresses")
finish_read_dir = time.time()
batch_size = int(args.bs)
confidence = float(args.confidence)
nms_thesh = float(args.nms_thresh)
namesfile = args.names
cuda_present = torch.cuda.is_available()
classes = load_classes(namesfile)
num_classes = len(classes)
# Set up the neural network
load_net = time.time()
print_info(widget, False, "info", "Loading network.....")
model = Darknet(args.cfg)
model.load_weights(args.weights)
print_info(widget, False, "info", "Network successfully loaded")
finish_load_net = time.time()
model.net_info["height"] = args.reso
model.net_info["width"] = 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_present:
model.cuda()
# Set the model in evaluation mode (for Batchnorm layers)
model.eval()
# Detection phase
load_batch = time.time()
print_info(widget, False, "info", "Loading batches.....")
loaded_ims = [cv2.imread(x) for x in im_list]
im_batches = list(map(prep_image, loaded_ims, [inp_dim for _ in range(len(im_list))]))
im_dim_list = [(x.shape[1], x.shape[0]) for x in loaded_ims]
im_dim_list = torch.FloatTensor(im_dim_list).repeat(1, 2)
leftover = 0
if len(im_dim_list) % batch_size:
leftover = 1
if batch_size != 1:
num_batches = len(im_list) // 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)]
if cuda_present:
im_dim_list = im_dim_list.cuda()
output = torch.empty((0, 0))
print_info(widget, False, "info", "Finished loading batches....")
start_det_loop = time.time()
for i, batch in enumerate(im_batches):
# load the image
start = time.time()
print_info(widget, False, "info", f"Detecting batch no {i}....")
if cuda_present:
batch = batch.cuda()
with torch.no_grad():
prediction = model(batch, cuda_present)
prediction = write_results(prediction, confidence, num_classes, nms_conf=nms_thesh)
end = time.time()
if type(prediction) == int:
for im_num, image in enumerate(im_list[i * batch_size: min((i + 1) * batch_size, len(im_list))]):
im_id = i * batch_size + im_num
msg = "{0:20s} predicted in {1:6.3f} seconds".format(image.split("/")[-1], (end - start) / batch_size)
msg += "\n{0:20s} {1:s}".format("Objects Detected:", "")
msg += "\n----------------------------------------------------------"
print_info(widget, False, 'batch_info', msg, im_id)
continue
prediction[:, 0] += i * batch_size # transform the atribute from index in batch to index in imlist
if np.size(output, 0) == 0: # If we have't initialised output
output = prediction
else:
output = torch.cat((output, prediction))
for im_num, image in enumerate(im_list[i * batch_size: min((i + 1) * batch_size, len(im_list))]):
im_id = i * batch_size + im_num
objs = [classes[int(x[-1])] for x in output if int(x[0]) == im_id]
msg = "{0:20s} predicted in {1:6.3f} seconds".format(image.split("/")[-1], (end - start) / batch_size)
msg += "\n{0:20s} {1:s}".format("Objects Detected:", " ".join(objs))
msg += "\n----------------------------------------------------------"
print_info(widget, False, 'batch_info', msg, im_id)
if cuda_present:
torch.cuda.synchronize()
print_info(widget, False, "info", f"Finished detecting batch no {i}")
if np.size(output, 0) == 0:
print_info(widget, False, 'no_detections', "No detections were made")
print_info(widget, False, 'finished')
return
# Start rescaling
print_info(widget, False, "info", "Output processing....")
output_rescale = time.time()
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
# set padding space black
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])
class_load = time.time()
print_info(widget, False, "info", "Finished output processing.")
# Start draw
print_info(widget, False, "info", "Drawing boxes....")
draw = time.time()
images_handler = ImagesHandler(classes, output, loaded_ims, args.det, im_list, batch_size)
images_handler.write()
print_info(widget, False, "images_ready", images_handler.imageList)
end = time.time()
print_info(widget, False, "info", "Finished drawing boxes")
msg = "\n\nSUMMARY"
msg += "\n----------------------------------------------------------"
msg += "\n{:25s}: {}".format("Task", "Time Taken (in seconds)")
msg += "\n"
msg += "\n{:25s}: {:2.3f}".format("Reading addresses", finish_read_dir - read_dir)
msg += "\n{:25s}: {:2.3f}".format("Loading network", finish_load_net - load_net)
msg += "\n{:25s}: {:2.3f}".format("Loading batch", start_det_loop - load_batch)
msg += "\n{:25s}: {:2.3f}".format("Detection (" + str(len(im_list)) + " images)", output_rescale - start_det_loop)
msg += "\n{:25s}: {:2.3f}".format("Output Processing", class_load - output_rescale)
msg += "\n{:25s}: {:2.3f}".format("Drawing Boxes", end - draw)
msg += "\n{:25s}: {:2.3f}".format("Average time_per_img", (end - load_batch) / len(im_list))
msg += "\n----------------------------------------------------------"
print_info(widget, False, 'info', msg)
torch.cuda.empty_cache()
print_info(widget, False, 'finished')
def train(params):
params = Params(params)
set_random_seeds(params.seed)
time_now = datetime.datetime.now().strftime("%Y%m%d%H%M%S")
params.save_root = params.save_root + f'/{params.project_name}_{time_now}_{params.version}'
os.makedirs(params.save_root, exist_ok=True)
logging.basicConfig(
filename=
f'{params.save_root}/{params.project_name}_{time_now}_{params.version}.log',
filemode='a',
format='%{asctime}s - %(levalname)s: %(message)s')
if params.num_gpus == 0:
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
logging.info(f'Available GPUs: {torch.cuda.device_count()}')
# Train pipeline
files = glob.glob(
os.path.join(params.data_root, params.project_name, params.train_set,
'*/*.JPEG'))
labels = []
for fp in files:
label = int(fp.split('/')[-2]) - 1
labels.append(label)
assert len(files) == len(labels)
train_pipeline = TrainImageDecoderPipeline(params=params,
device_id=0,
files=files,
labels=labels)
train_pipeline.build()
train_pii = pytorchIterator(train_pipeline,
last_batch_policy=LastBatchPolicy.DROP,
reader_name='Reader',
auto_reset=True)
# Evaluation pipeline
files = glob.glob(
os.path.join(params.data_root, params.project_name, params.val_set,
'*.JPEG'))
files = sorted(files,
key=lambda f: f.split('/')[-1].split('_')[-1].split('.')[0])
labels = loadlabel(
os.path.join(
params.data_root, params.project_name,
'ILSVRC2012_devkit_t12/data/ILSVRC2012_validation_ground_truth.txt'
))
eval_pipeline = EvalImageDecoderPipeline(params=params,
device_id=0,
files=files,
labels=labels)
eval_pipeline.build()
eval_pii = pytorchIterator(eval_pipeline,
last_batch_policy=LastBatchPolicy.PARTIAL,
reader_name='Reader',
auto_reset=True)
model = Darknet()
last_step = 0
last_epoch = 0
if params.load_weights != 'None':
try:
state_dict = torch.load(params.load_weights)
model.load_state_dict(state_dict)
last_step = int(params.load_weights.split('_')[-1].split('.')[0])
last_epoch = int(params.load_weights.split('_')[-2])
except:
logging.error('Fail to resuming from weight!')
exit()
if params.num_gpus > 0:
model = model.cuda()
if params.num_gpus > 1:
model = nn.DataParallel(model)
if params.optim == 'Adam':
optimizer = torch.optim.Adam(model.parameters(),
lr=params.learning_rate)
else:
optimizer = torch.optim.SGD(model.parameters(),
lr=params.learning_rate,
momentum=0.9,
nesterov=True)
criterion = nn.CrossEntropyLoss()
# ls_scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(optimizer=optimizer, factor=0.5, verbose=True, patience=8)
epoch = 0
begin_epoch = max(0, last_epoch)
step = max(0, last_step)
best_loss = 100
logging.info('Begin to train...')
model.train()
try:
for epoch in range(begin_epoch, params.epoch):
for iter, data in enumerate(train_pii):
x = data[0]['data']
label = data[0]['label'].squeeze(-1).long().cuda()
output = model(x)
loss = criterion(output, label)
optimizer.zero_grad()
loss.backward()
optimizer.step()
if iter % params.save_interval == 0:
logging.info(
f'{datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")} '
f'Train Epoch: {epoch} iter: {iter} loss: {loss.item()}'
)
step += 1
if epoch % params.eval_interval == 0:
model.eval()
epoch_loss = 0
prediciton = []
target = []
with torch.no_grad():
for iter, data in enumerate(eval_pii):
x = data[0]['data']
label = data[0]['label'].squeeze(-1).long().cuda()
output = model(x)
loss = criterion(output, label).item()
epoch_loss += loss * x.shape[0]
prediciton.append(output)
target.append(label)
loss = epoch_loss / 50000
prediciton = torch.cat(prediciton, dim=0)
target = torch.cat(target, dim=0)
acc = top1accuracy(prediciton, target)
acctop5 = top5accuracy(prediciton, target)
logging.info(
f'{datetime.datetime.now().strftime("%Y-%m-%d %H:%M:%S")} '
f'Eval Epoch: {epoch} loss: {loss} accuracy: {acc} Top5 acc: {acctop5}'
)
if loss < best_loss:
best_loss = loss
save_checkpoint(
model, f'{params.save_root}/{epoch}_{step}.pth')
model.train()
except KeyboardInterrupt:
save_checkpoint(model,
f'{params.save_root}/Interrupt_{epoch}_{step}.pth')
help="the image to predict (default: %(default)s)")
parser.add_argument("--weight", required=True, metavar="/path/to/yolov4.weights", help="the path of weight file")
parser.add_argument("--save-img", metavar="predicted-img", help="the path to save predicted image")
args = parser.parse_args()
return args
if __name__ == "__main__":
args = parse_args()
img: Image.Image = Image.open(args.img_file)
img = img.resize((608, 608))
# C*H*W
img_data = to_image(img)
net = Darknet(img_data.size(0))
net.load_weights(args.weight)
net.eval()
with torch.no_grad():
boxes, confs = net(img_data.unsqueeze(0))
idxes_pred, boxes_pred, probs_pred = utils.post_processing(boxes, confs, 0.4, 0.6)
utils.plot_box(boxes_pred, args.img_file, args.save_img)
def detect(kitti_weights='../checkpoints/best_weights_kitti.pth', config_path='../config/yolov3-kitti.cfg',
class_path='../data/names.txt'):
"""
Script to run inference on sample images. It will store all the inference results in /output directory (
relative to repo root)
Args
kitti_weights: Path of weights
config_path: Yolo configuration file path
class_path: Path of class names txt file
"""
cuda = torch.cuda.is_available()
os.makedirs('../output', exist_ok=True)
# Set up model
model = Darknet(config_path, img_size=416)
model.load_weights(kitti_weights)
if cuda:
model.cuda()
print("Cuda available for inference")
model.eval() # Set in evaluation mode
dataloader = DataLoader(ImageFolder("../data/samples/", img_size=416),
batch_size=2, shuffle=False, num_workers=0)
classes = load_classes(class_path) # Extracts class labels from file
Tensor = torch.cuda.FloatTensor if cuda else torch.FloatTensor
imgs = [] # Stores image paths
img_detections = [] # Stores detections for each image index
print('data size : %d' % len(dataloader))
print('\nPerforming object detection:')
prev_time = time.time()
for batch_i, (img_paths, input_imgs) in enumerate(dataloader):
# Configure input
input_imgs = Variable(input_imgs.type(Tensor))
# Get detections
with torch.no_grad():
detections = model(input_imgs)
detections = non_max_suppression(detections, 80, 0.8, 0.4)
# print(detections)
# Log progress
current_time = time.time()
inference_time = datetime.timedelta(seconds=current_time - prev_time)
prev_time = current_time
print('\t+ Batch %d, Inference Time: %s' % (batch_i, inference_time))
# Save image and detections
imgs.extend(img_paths)
img_detections.extend(detections)
# Bounding-box colors
# cmap = plt.get_cmap('tab20b')
cmap = plt.get_cmap('tab10')
colors = [cmap(i) for i in np.linspace(0, 1, 20)]
print('\nSaving images:')
# Iterate through images and save plot of detections
for img_i, (path, detections) in enumerate(zip(imgs, img_detections)):
print("(%d) Image: '%s'" % (img_i, path))
# Create plot
img = np.array(Image.open(path))
plt.figure()
fig, ax = plt.subplots(1)
ax.imshow(img)
kitti_img_size = 416
# The amount of padding that was added
pad_x = max(img.shape[0] - img.shape[1], 0) * (kitti_img_size / max(img.shape))
pad_y = max(img.shape[1] - img.shape[0], 0) * (kitti_img_size / max(img.shape))
# Image height and width after padding is removed
unpad_h = kitti_img_size - pad_y
unpad_w = kitti_img_size - pad_x
# Draw bounding boxes and labels of detections
if detections is not None:
print(type(detections))
print(detections.size())
unique_labels = detections[:, -1].cpu().unique()
n_cls_preds = len(unique_labels)
bbox_colors = random.sample(colors, n_cls_preds)
for x1, y1, x2, y2, conf, cls_conf, cls_pred in detections:
print('\t+ Label: %s, Conf: %.5f' % (classes[int(cls_pred)], cls_conf.item()))
# Rescale coordinates to original dimensions
box_h = int(((y2 - y1) / unpad_h) * (img.shape[0]))
box_w = int(((x2 - x1) / unpad_w) * (img.shape[1]))
y1 = int(((y1 - pad_y // 2) / unpad_h) * (img.shape[0]))
x1 = int(((x1 - pad_x // 2) / unpad_w) * (img.shape[1]))
color = bbox_colors[int(np.where(unique_labels == int(cls_pred))[0])]
# Create a Rectangle patch
bbox = patches.Rectangle((x1, y1), box_w, box_h, linewidth=2,
edgecolor=color,
facecolor='none')
# Add the bbox to the plot
ax.add_patch(bbox)
# Add label
plt.text(x1, y1 - 30, s=classes[int(cls_pred)] + ' ' + str('%.4f' % cls_conf.item()), color='white',
verticalalignment='top',
bbox={'color': color, 'pad': 0})
# Save generated image with detections
plt.axis('off')
plt.gca().xaxis.set_major_locator(NullLocator())
plt.gca().yaxis.set_major_locator(NullLocator())
plt.savefig('../output/%d.png' % (img_i), bbox_inches='tight', pad_inches=0.0)
plt.close()
class ReadFramesThread:
def __init__(self, path, args, with_tracking, widget, queue_size=3000):
# initialize the file video stream along with the boolean
# used to indicate if the thread should be stopped or not
self.stream = cv2.VideoCapture(path)
self.widget = widget
self.tracking = with_tracking
if not self.stream:
if type(path) == int:
print_info(widget, True, "error",
f"Error opening web cam on {path}")
else:
print_info(widget, True, "error",
f"Error opening video file {path}")
self.stopped = False
self.canceled = False
self.paused = False
self.ready = False
# initialize the queue used to store frames read from
# the video file
self.Q = Queue(maxsize=queue_size)
self.imread = Queue(maxsize=queue_size)
self.Q_processed = Queue(maxsize=queue_size)
self.inp_dim = int(args.reso)
self.batch_size = int(args.bs)
self.names_file = args.names
self.confidence = float(args.confidence)
self.nms_thresh = float(args.nms_thresh)
self.is_classifier = args.is_classifier
self.classes = load_classes(self.names_file)
self.num_classes = len(self.classes)
self.model = None
self.model_classifier = None
if self.is_classifier:
print_info(widget, False, "info",
"Loading network for detection.....", -1)
self.model = Darknet(args.classifier_cfg)
self.model.load_weights(args.classifier_weights)
print_info(widget, False, "info",
"Network for detection successfully loaded", 0)
print_info(widget, False, "info",
"Loading network for classification.....", -1)
self.model_classifier = Darknet(args.cfg)
self.model_classifier.load_weights(args.weights)
print_info(widget, False, "info",
"Network for classification successfully loaded", 0)
self.model_classifier.net_info["height"] = args.reso
self.inp_dim = int(self.model_classifier.net_info["height"])
# If there's a GPU availible, put the model on GPU
self.cuda = torch.cuda.is_available()
if self.cuda:
self.model_classifier.cuda()
# Set the model in evaluation mode
self.model_classifier.eval()
self.classifier_confidence = self.confidence
self.classifier_nms_thesh = self.nms_thresh
self.classifier_classes = self.classes
self.classifier_num_classes = self.num_classes
self.classifier_names_file = self.names_file
self.classifier_inp_dim = self.inp_dim
self.inp_dim = args.classifier_inp_dim
self.confidence = args.classifier_confidence
self.nms_thresh = args.classifier_nms_thresh
self.names_file = args.classifier_names
self.classes = load_classes(self.names_file)
self.num_classes = len(self.classes)
else:
print_info(widget, False, "info", "Loading network.....", -1)
self.model = Darknet(args.cfg)
self.model.load_weights(args.weights)
print_info(widget, False, "info", "Network successfully loaded", 0)
self.model.net_info["height"] = self.inp_dim
assert self.inp_dim % 32 == 0
assert self.inp_dim > 32
# If there's a GPU availible, put the model on GPU
self.cuda = torch.cuda.is_available()
if self.cuda:
self.model.cuda()
# Set the model in evaluation mode
self.model.eval()
# if tracking selected, initialize sort class
self.mot_tracking = None
if self.tracking == "sort":
self.mot_tracking = Sort(max_age=30, min_hits=3)
elif self.tracking == "deep_sort":
print_info(widget, False, "info", "Loading Deep Sort model ...",
-1)
self.mot_tracking = DeepSort()
print_info(widget, False, "info", "Deep Sort model loaded", -1)
def start(self):
# start a thread to read frames from the file video stream
t = Thread(target=self.update, args=())
# t.daemon = True
t.start()
return self
def update(self):
frames = 0
start = time.time()
print_info(self.widget, False, "info", "Began capturing", -2)
# keep looping infinitely
while True:
# if the thread indicator variable is set, stop the
# thread
if self.stopped:
break
if self.canceled:
current_time = time.time()
print_info(self.widget, False, "info", "Canceled processing",
current_time - start)
return
if self.paused:
self.widget.obj.pauseMutex.lock()
self.widget.obj.pauseCond.wait(self.widget.obj.pauseMutex)
self.widget.obj.pauseMutex.unlock()
self.paused = False
# otherwise, ensure the queue has room in it
if not self.Q.full():
# read the next frame from the file
(grabbed, frame) = self.stream.read()
# if the `grabbed` boolean is `False`, then we have
# reached the end of the video file
if not grabbed:
self.stop()
self.ready = True
return
# add the frame to the queue
self.Q.put(prep_image(frame, self.inp_dim))
self.imread.put(frame)
frames += 1
current_time = time.time()
msg = " FPS of the video is {:5.4f}".format(
frames / (current_time - start))
print_info(self.widget, False, "info", msg,
current_time - start)
if frames % self.batch_size == 0:
self.process_frames()
if not self.Q.empty():
self.process_frames()
def read(self):
# return next frame in the queue
return self.Q.get()
def more(self):
# return True if there are still frames in the queue
return self.Q.qsize() > 0
def stop(self):
# indicate that the thread should be stopped
self.stopped = True
def cancel(self):
self.canceled = True
def pause(self):
self.paused = True
def has_batch(self):
if self.Q.qsize() >= self.batch_size:
return True
if self.Q.qsize() > 0 and self.stopped:
return True
return False
def get_batch(self):
if (self.Q.qsize() >= self.batch_size) or (self.Q.qsize() > 0
and self.stopped):
res = np.empty((0, 0))
im_dim_list = []
imread_list = []
for _ in range(self.batch_size):
img = self.Q.get()
if np.size(res, 0) == 0:
res = img
else:
res = torch.cat((res, img))
img = self.imread.get()
im_dim_list.append((img.shape[1], img.shape[0]))
imread_list.append(img)
im_dim_list = torch.FloatTensor(im_dim_list).repeat(1, 2)
return res, im_dim_list, imread_list
return False, False, False
def process_frames(self):
batch_nr = -1
batch, im_dims, imread = self.get_batch()
if imread:
batch_nr += 1
if self.cuda:
im_dims = im_dims.cuda()
batch = batch.cuda()
with torch.no_grad():
output = self.model(batch, self.cuda)
for frame_id in range(np.size(output, 0)):
nr_frame = self.batch_size * batch_nr + frame_id + 1
im_dim = im_dims[frame_id]
frame = output[frame_id].unsqueeze(0)
frame = write_results(frame,
self.confidence,
self.num_classes,
nms_conf=self.nms_thresh)
if np.size(frame, 0) > 0:
im_dim = im_dim.repeat(frame.size(0), 1)
scaling_factor = torch.min(416 / im_dim, 1)[0].view(-1, 1)
frame[:, [1, 3]] -= (self.inp_dim - scaling_factor *
im_dim[:, 0].view(-1, 1)) / 2
frame[:, [2, 4]] -= (self.inp_dim - scaling_factor *
im_dim[:, 1].view(-1, 1)) / 2
frame[:, 1:5] /= scaling_factor
for i in range(frame.shape[0]):
frame[i, [1, 3]] = torch.clamp(frame[i, [1, 3]], 0.0,
im_dim[i, 0])
frame[i, [2, 4]] = torch.clamp(frame[i, [2, 4]], 0.0,
im_dim[i, 1])
if self.is_classifier:
frame = self.apply_classifier_model(
imread[frame_id], frame)
if self.tracking == "sort":
if self.cuda:
frame = frame.cpu()
frame = self.mot_tracking.update(frame)
if self.cuda:
frame = torch.from_numpy(frame).cuda()
elif self.tracking == "deep_sort":
if self.cuda:
frame = frame.cpu()
tracker, detections_class = self.mot_tracking.update(
imread[frame_id], frame)
frame = []
for track in tracker.tracks:
if not track.is_confirmed(
) or track.time_since_update > 1:
continue
bbox = track.to_tlbr(
) # Get the corrected/predicted bounding box
id_num = int(track.track_id
) # Get the ID for the particular track.
# Draw bbox from tracker.
frame.append(
np.concatenate(([id_num + 1], bbox, [
track.conf_score, track.class_score, track.cid
])).reshape(1, -1))
if len(frame) > 0:
frame = np.concatenate(frame)
if self.cuda:
frame = torch.from_numpy(frame).cuda()
else:
frame = torch.empty((0, 8))
if np.size(frame, 0) == 0:
image_handler = ImageHandler(nr_frame, batch_nr,
f"frame{nr_frame}",
imread[frame_id],
self.tracking)
self.Q_processed.put(image_handler)
continue
image_handler = ImageHandler(nr_frame, batch_nr,
f"frame{nr_frame}",
imread[frame_id], self.tracking)
if self.is_classifier:
image_handler.write(frame, self.classifier_classes)
else:
image_handler.write(frame, self.classes)
self.Q_processed.put(image_handler)
def get_image(self):
return self.Q_processed.get()
def has_images(self):
return not self.Q_processed.empty()
def apply_classifier_model(self, imread, frame):
# get crops from detections in frame
crops = torch.empty((0, 0))
detections = frame[:, 1:5]
for d in detections:
for i in range(len(d)):
if d[i] < 0:
d[i] = 0
img_h, img_w, img_ch = imread.shape
xmin, ymin, xmax, ymax = d
if xmin > img_w:
xmin = img_w
if ymin > img_h:
ymin = img_h
ymin = abs(int(ymin))
ymax = abs(int(ymax))
xmin = abs(int(xmin))
xmax = abs(int(xmax))
try:
crop = imread[ymin:ymax, xmin:xmax, :]
crop = prep_image(crop, self.classifier_inp_dim)
if np.size(crops, 0) == 0:
crops = crop
else:
crops = torch.cat((crops, crop))
except:
continue
if self.cuda:
crops = crops.cuda()
with torch.no_grad():
output = self.model_classifier(crops, self.cuda)
for frame_id in range(np.size(output, 0)):
new_det = output[frame_id].unsqueeze(0)
new_det = write_results(new_det,
self.classifier_confidence,
self.classifier_num_classes,
nms_conf=self.classifier_nms_thesh)
if np.size(new_det, 0) > 0:
index = torch.argmax(new_det[:, 6])
frame[frame_id, 6:8] = new_det[index, 6:8]
else:
frame[frame_id, 6] = -1
frame = frame[frame[:, 6] >= 0]
return frame
def infer(payload):
unlabeled = payload["unlabeled"]
ckpt_file = payload["ckpt_file"]
batch_size = 16
coco = COCO("./data", Transforms(), samples=unlabeled, train=True)
loader = DataLoader(coco,
shuffle=False,
batch_size=batch_size,
collate_fn=collate_fn)
config_file = "yolov3.cfg"
model = Darknet(config_file).to(device)
ckpt = torch.load(os.path.join("./log", ckpt_file))
model.load_state_dict(ckpt["model"])
model.eval()
# batch predictions from the entire test set
predictions = []
with torch.no_grad():
for img, _, _ in loader:
img = img.to(device)
# get inference output
output = model(img)
# batch predictions from 3 yolo layers
batched_prediction = []
for p in output: # (batch_size, 3, gx, gy, 85)
batch_size = p.shape[0]
p = p.view(batch_size, -1, 85)
batched_prediction.append(p)
batched_prediction = torch.cat(batched_prediction, dim=1)
predictions.append(batched_prediction)
predictions = torch.cat(predictions, dim=0)
# apply nms to predicted bounding boxes
predicted_boxes, predicted_objectness, predicted_class_dist = bbox_transform(
predictions)
# the predicted boxes are in log space relative to the anchor priors
# bring them back to normalized xyxy format
cxcy_priors = anchors.normalize("cxcy")
# expand the priors to match the dimension of predicted_boxes
batched_cxcy_priors = cxcy_priors.unsqueeze(0).repeat(
predicted_boxes.shape[0], 1, 1)
predicted_boxes = batched_gcxgcy_to_cxcy(predicted_boxes,
batched_cxcy_priors)
del batched_cxcy_priors
# convert predicted_boxes to xyxy format and perform nms
xyxy = batched_cxcy_to_xy(predicted_boxes)
del predicted_boxes # (no longer need cxcy format)
# class distribution is part of the return
# do notapply softmax to the predicted class distribution
# as we will do it internally for efficiency
outputs = {}
for i in range(len(coco)):
# get boxes, scores, and objects on each image
_xyxy, _scores = xyxy[i], predicted_objectness[i]
_pre_softmax = predicted_class_dist[i]
keep = tv.ops.nms(_xyxy, _scores, 0.5)
boxes, scores, pre_softmax = _xyxy[keep], _scores[keep], _pre_softmax[
keep]
outputs[i] = {
"boxes": boxes.cpu().numpy().tolist(),
"pre_softmax": pre_softmax.cpu().numpy().tolist(),
"scores": scores.cpu().numpy().tolist(),
}
return {"outputs": outputs}
def test(payload):
ckpt_file = payload["ckpt_file"]
batch_size = 16
coco = COCO("./data", Transforms(), train=False)
loader = DataLoader(coco,
shuffle=False,
batch_size=batch_size,
collate_fn=collate_fn)
config_file = "yolov3.cfg"
model = Darknet(config_file).to(device)
ckpt = torch.load(os.path.join("./log", ckpt_file))
model.load_state_dict(ckpt["model"])
model.eval()
# batch predictions from the entire test set
predictions = []
# keep track of ground-truth boxes and label
labels = []
with torch.no_grad():
for img, boxes, class_labels in loader:
img = img.to(device)
# get inference output
output = model(img)
for b, c in zip(boxes, class_labels):
labels.append((b, c))
# batch predictions from 3 yolo layers
batched_prediction = []
for p in output: # (bacth_size, 3, gx, gy, 85)
p = p.view(p.shape[0], -1, 85)
batched_prediction.append(p)
batched_prediction = torch.cat(batched_prediction, dim=1)
predictions.append(batched_prediction)
predictions = torch.cat(predictions, dim=0)
# apply nms to predicted bounding boxes
predicted_boxes, predicted_objectness, predicted_class_dist = bbox_transform(
predictions)
# the predicted boxes are in log space relative to the anchor priors
# bring them back to normalized xyxy format
cxcy_priors = anchors.normalize("cxcy")
# expand the priors to match the dimension of predicted_boxes
batched_cxcy_priors = cxcy_priors.unsqueeze(0).repeat(
predicted_boxes.shape[0], 1, 1)
predicted_boxes = batched_gcxgcy_to_cxcy(predicted_boxes,
batched_cxcy_priors)
del batched_cxcy_priors
# convert predicted_boxes to xyxy format and perform nms
xyxy = batched_cxcy_to_xy(predicted_boxes)
del predicted_boxes # (no longer need cxcy format)
# get predicted object
# apply softmax to the predicted class distribution
# note that bbox_tranform does not apply softmax
# because the loss we are using requires us to use raw output
predicted_objects = torch.argmax(F.softmax(predicted_class_dist, dim=-1),
dim=-1)
# predictions on the test set (value of "predictions" of the return)
prd = {}
for i in range(len(coco)):
# get boxes, scores, and objects on each image
_xyxy, _scores = xyxy[i], predicted_objectness[i]
_objects = predicted_objects[i]
keep = tv.ops.nms(_xyxy, _scores, 0.5)
boxes, scores, objects = _xyxy[keep], _scores[keep], _objects[keep]
prd[i] = {
"boxes": boxes.cpu().numpy().tolist(),
"objects": objects.cpu().numpy().tolist(),
"scores": scores.cpu().numpy().tolist(),
}
# ground-truth of the test set
# skip "difficulties" field, because every object in COCO
# should be considered reasonable
lbs = {}
for i in range(len(coco)):
boxes, class_labels = labels[i]
lbs[i] = {
"boxes": boxes.cpu().numpy().tolist(),
"objects": class_labels
}
return {"predictions": prd, "labels": lbs}
