def val():
model.eval()
mean = np.array([0.485, 0.456, 0.406],
dtype=np.float32).reshape(1, 1, 3)
std = np.array([0.229, 0.224, 0.225],
dtype=np.float32).reshape(1, 1, 3)
t3 = time.time()
res = []
i = 1
# file_lists = sorted(glob.glob(args.img_list))
file_lists = open('val.txt')
for file in tqdm(file_lists.readlines()):
torch.cuda.synchronize()
seg = file[:-2].split('_')
img = plt.imread(
'/mnt/raid/Talor/CSP_res2net/data/citypersons/images/val/'
+ seg[0] + '/' + file[:-2]).astype(np.float32)
img_pre = preprocess(img[:, :, ::-1], mean, std)
img_pre = img_pre.cuda()
with torch.no_grad():
output = model(img_pre)[-1]
output['hm'].sigmoid_()
hm, wh, reg, attr = output['hm'], output['wh'], output[
'reg'], output['aed']
density = attr.pow(2).sum(dim=1, keepdim=True).sqrt()
diversity = torch.div(attr, density)
boxes = parse_det(hm,
wh,
reg,
density=density,
diversity=diversity,
score=0.01,
down=4)
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
for box in boxes:
temp = dict()
temp['image_id'] = i
temp['category_id'] = 1
temp['bbox'] = box[:4].tolist()
temp['score'] = float(box[4])
res.append(temp)
i = i + 1
with open('./_temp_val.json', 'w') as f:
json.dump(res, f)
MRs = validate('./eval_city/val_gt.json', './_temp_val.json')
t4 = time.time()
print(
'Summerize: [Reasonable: %.2f%%], [Bare: %.2f%%], [Partial: %.2f%%], [Heavy: %.2f%%]'
% (MRs[0] * 100, MRs[1] * 100, MRs[2] * 100, MRs[3] * 100))
print('Validation time used: %.3f' % (t4 - t3))
return MRs[0]
def val(log=None):
net.eval()
if config.teacher:
print('Load teacher params')
student_dict = net.module.state_dict()
net.module.load_state_dict(teacher_dict)
print('Perform validation...')
res = []
t3 = time.time()
for i, data in enumerate(testloader, 0):
inputs = data.cuda()
with torch.no_grad():
pos, height, offset = net(inputs)
boxes = parse_det_offset(pos.cpu().numpy(),
height.cpu().numpy(),
offset.cpu().numpy(),
config.size_test,
score=0.1,
down=4,
nms_thresh=0.5)
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
for box in boxes:
temp = dict()
temp['image_id'] = i + 1
temp['category_id'] = 1
temp['bbox'] = box[:4].tolist()
temp['score'] = float(box[4])
res.append(temp)
print('\r%d/%d' % (i + 1, len(testloader))),
sys.stdout.flush()
print('')
if config.teacher:
print('Load back student params')
net.module.load_state_dict(student_dict)
with open('./_temp_val.json', 'w') as f:
json.dump(res, f)
MRs = validate('./eval_city/val_gt.json', './_temp_val.json')
t4 = time.time()
print(
'Summerize: [Reasonable: %.2f%%], [Bare: %.2f%%], [Partial: %.2f%%], [Heavy: %.2f%%]'
% (MRs[0] * 100, MRs[1] * 100, MRs[2] * 100, MRs[3] * 100))
if log is not None:
log.write("%.7f %.7f %.7f %.7f\n" % tuple(MRs))
print('Validation time used: %.3f' % (t4 - t3))
return MRs[0]
def val(testloader, net, config, args, teacher_dict=None):
net.eval()
print('Perform validation...')
res = []
inference_time = 0
num_images = len(testloader)
for i, data in enumerate(testloader):
inputs = data.cuda()
with torch.no_grad():
t1 = time.time()
pos, height, offset = net(inputs)
t2 = time.time()
inference_time += (t2 - t1)
boxes = parse_det_offset(pos.cpu().numpy(),
height.cpu().numpy(),
offset.cpu().numpy(),
config.size_test,
score=0.1,
down=4,
nms_thresh=0.5)
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
for box in boxes:
temp = dict()
temp['image_id'] = i + 1
temp['category_id'] = 1
temp['bbox'] = box[:4].tolist()
temp['score'] = float(box[4])
res.append(temp)
print('\r%d/%d' % (i + 1, num_images), end='')
sys.stdout.flush()
temp_val = args.json_out
with open(temp_val, 'w') as f:
json.dump(res, f)
MRs = validate('./eval_city/val_gt.json', temp_val)
print(
'\nSummerize:[Reasonable: %.2f%%], [Reasonable_small: %.2f%%], [Reasonable_occ=heavy: %.2f%%], [All: %.2f%%]'
% (MRs[0] * 100, MRs[1] * 100, MRs[2] * 100, MRs[3] * 100))
FPS = int(num_images / inference_time)
print('FPS : {}'.format(FPS))
return MRs
def val(log=None):
net1.eval()
model.eval()
# net.eval()
print('Perform validation...')
res = []
t3 = time.time()
for i, data in enumerate(testloader, 0):
tt1 = time.time()
inputs = data[0].cuda()
img = plt.imread(data[1][0]).astype(np.float32)
img_pre = preprocess(img[:, :, ::-1], mean, std)
img_pre = img_pre.cuda()
with torch.no_grad():
pos1, height1, offset1 = net1(inputs)
# pos, height, offset = net(inputs)
output = model(img_pre)[-1]
output['hm'].sigmoid_()
hm, wh, reg, attr = output['hm'], output['wh'], output['reg'], output[
'aed']
density = attr.pow(2).sum(dim=1, keepdim=True).sqrt()
diversity = torch.div(attr, density)
boxes = parse_det(hm,
wh,
reg,
density=density,
diversity=diversity,
score=0.01,
down=4)
# boxes = parse_det_offset(pos.cpu().numpy(), height.cpu().numpy(), offset.cpu().numpy(), config.size_test, score=0.01, down=4, nms_thresh=0.5)
boxes1 = parse_det_offset(pos1.cpu().numpy(),
height1.cpu().numpy(),
offset1.cpu().numpy(),
config.size_test,
score=0.01,
down=4,
nms_thresh=0.5)
bb = list(boxes) + list(boxes1)
# bb = list(boxes)
boxes = np.asarray(bb, dtype=np.float32)
keep = nms(boxes, 0.5, usegpu=False, gpu_id=0)
boxes = boxes[keep, :]
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
for box in boxes:
temp = dict()
temp['image_id'] = i + 1
temp['category_id'] = 1
temp['bbox'] = box[:4].tolist()
temp['score'] = float(box[4])
res.append(temp)
tt2 = time.time()
print('\r%d/%d, %f' % (i + 1, len(testloader), tt2 - tt1)),
sys.stdout.flush()
print('')
if config.teacher:
print('Load back student params')
net.module.load_state_dict(student_dict)
with open('./_temp_val.json', 'w') as f:
json.dump(res, f)
MRs = validate('./eval_city/val_gt.json', './_temp_val.json')
t4 = time.time()
print(
'Summerize: [Reasonable: %.2f%%], [Bare: %.2f%%], [Partial: %.2f%%], [Heavy: %.2f%%]'
% (MRs[0] * 100, MRs[1] * 100, MRs[2] * 100, MRs[3] * 100))
if log is not None:
log.write("%.7f %.7f %.7f %.7f\n" % tuple(MRs))
print('Validation time used: %.3f' % (t4 - t3))
return MRs[0]
def main():
# BGR
mean = np.array([0.485, 0.456, 0.406],
dtype=np.float32).reshape(1, 1, 3)
std = np.array([0.229, 0.224, 0.225],
dtype=np.float32).reshape(1, 1, 3)
args = parse_args()
num_layers = 34
heads = {'hm': 1, 'wh': 1, 'reg': 2, 'aed': 4}
model = peddla_net(num_layers, heads, head_conv=256, down_ratio=4).cuda().eval()
# load model
# model = load_model(model, 'final.pth')
model.load_state_dict(torch.load('./ckpt/CSID-5.pth'))
# torch.cuda.empty_cache()
t3 = time.time()
res = []
i = 1
# file_lists = sorted(glob.glob(args.img_list))
file_lists = open('val.txt')
for file in tqdm(file_lists.readlines()):
torch.cuda.synchronize()
seg = file[:-2].split('_')
img = plt.imread('/mnt/raid/Talor/CSP_res2net/data/citypersons/images/val/' + seg[0] + '/' + file[:-2]).astype(np.float32)
img_pre = preprocess(img[:, :, ::-1], mean, std)
img_pre = img_pre.cuda()
with torch.no_grad():
output = model(img_pre)[-1]
output['hm'].sigmoid_()
hm, wh, reg, attr = output['hm'], output['wh'], output['reg'], output['aed']
density = attr.pow(2).sum(dim=1, keepdim=True).sqrt()
diversity = torch.div(attr, density)
# print(hm.shape, wh.shape, reg.shape, attr.shape, density.shape, diversity.shape)
boxes = parse_det(hm, wh, reg, density=density, diversity=diversity, score=0.01, down=4)
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
for box in boxes:
temp = dict()
temp['image_id'] = i
temp['category_id'] = 1
temp['bbox'] = box[:4].tolist()
temp['score'] = float(box[4])
res.append(temp)
# print('\r%d/%d' % (i, 500))
# sys.stdout.flush()
i = i + 1
'''
if len(boxes) > 0:
boxes[:, [2, 3]] -= boxes[:, [0, 1]]
fig, ax = plt.subplots(1)
ax.imshow(img)
for i in range(len(boxes)):
x, y, w, h, score = boxes[i]
rect = patches.Rectangle((x, y), w, h, linewidth=1, edgecolor='r', facecolor='none')
ax.add_patch(rect)
else:
plt.imshow(img)
plt.show()
'''
with open('./_temp_val.json', 'w') as f:
json.dump(res, f)
MRs = validate('./eval_city/val_gt.json', './_temp_val.json')
t4 = time.time()
print('Summerize: [Reasonable: %.2f%%], [Bare: %.2f%%], [Partial: %.2f%%], [Heavy: %.2f%%]'
% (MRs[0]*100, MRs[1]*100, MRs[2]*100, MRs[3]*100))
print('Validation time used: %.3f' % (t4 - t3))
return MRs[0]