def main():
# load config
dataset_root = '/media/zxy/Samsung_T5/Data/DataSets/LaSOT/LaSOT_test'
# create tracker
'''Pytracking-RF tracker'''
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
'''Refinement module'''
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34_15sr_fcn/SEcmnet_ep0040-a.pth.tar" # RF_CrsM_R34SR15FCN_a
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34/SEcmnet_ep0040-a.pth.tar" # dimp_dimp50RF_CrsM_R34SR20FCN_a-0_1
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34/SEcmnet_ep0040-b.pth.tar"
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34/SEcmnet_ep0040-c.pth.tar"
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34/SEcmnet_ep0040-d.pth.tar"
# refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEx_beta/SEcm_r34/SEcmnet_ep0040-e.pth.tar"
refine_path = "/home/zxy/Desktop/AlphaRefine/experiments/SEbcm/SEbcm-8gpu/SEbcmnet_ep0040.pth.tar" # RF_CrsM_ARv1_d
selector_path = 1
branches = ['corner', 'mask'][0:1]
sr = 2.0
input_sz = int(128 * sr) # 2.0 by default
RF_module = RefineModule(refine_path,
selector_path,
branches=branches,
search_factor=sr,
input_sz=input_sz)
RF_type = 'RF_CrsM_R34SR20_e'
# RF_type = 'RF_CrsM_ARv1'
model_name = args.tracker_name + '_' + args.tracker_param + '{}-{}'.format(
RF_type, selector_path) + '_%d' % (args.run_id)
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root,
load_img=False)
if args.dataset in ['VOT2016', 'VOT2018', 'VOT2019']:
total_lost = 0
# restart tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
frame_counter = 0
lost_number = 0
toc = 0
'''对refinement module计时'''
toc_refine = 0
pred_bboxes = []
for idx, (img, gt_bbox) in enumerate(video):
if len(gt_bbox) == 4:
gt_bbox = [
gt_bbox[0], gt_bbox[1], gt_bbox[0],
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1,
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1, gt_bbox[1]
]
tic = cv2.getTickCount()
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
if idx == frame_counter:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(
gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
'''##### initilize refinement module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
pred_bboxes.append(1)
elif idx > frame_counter:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
'''##### refine tracking results #####'''
result_dict = RF_module.refine(img_RGB,
np.array(pred_bbox))
bbox_report = result_dict['bbox_report']
bbox_state = result_dict['bbox_state']
'''report result and update state'''
pred_bbox = bbox_report
x1, y1, w, h = bbox_state.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
new_pos = torch.from_numpy(
np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(
np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() /
tracker.base_target_sz.prod())
##### update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
overlap = vot_overlap(pred_bbox, gt_bbox,
(img.shape[1], img.shape[0]))
if overlap > 0:
# not lost
pred_bboxes.append(pred_bbox)
else:
# lost object
pred_bboxes.append(2)
frame_counter = idx + 5 # skip 5 frames
lost_number += 1
else:
pred_bboxes.append(0)
toc += cv2.getTickCount() - tic
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > frame_counter:
cv2.polylines(
img, [np.array(gt_bbox, np.int).reshape(
(-1, 1, 2))], True, (0, 255, 0), 3)
if len(pred_bbox) == 8:
cv2.polylines(
img,
[np.array(pred_bbox, np.int).reshape(
(-1, 1, 2))], True, (0, 255, 255), 3)
else:
bbox = list(map(int, pred_bbox))
cv2.rectangle(img, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.putText(img, str(lost_number), (40, 80),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
video_path = os.path.join(save_dir, args.dataset, model_name,
'baseline', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
if isinstance(x, int):
f.write("{:d}\n".format(x))
else:
f.write(','.join([vot_float2str("%.4f", i)
for i in x]) + '\n')
print(
'({:3d}) Video: {:12s} Time: {:4.1f}s Speed: {:3.1f}fps Lost: {:d}'
.format(v_idx + 1, video.name, toc, idx / toc, lost_number))
total_lost += lost_number
print("{:s} total lost: {:d}".format(model_name, total_lost))
else:
# OPE tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(
gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
'''##### initilize refinement module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
if 'VOT2018-LT' == args.dataset:
pred_bboxes.append([1])
else:
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
'''##### refine tracking results #####'''
pred_bbox = RF_module.refine(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
new_pos = torch.from_numpy(
np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(
np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() /
tracker.base_target_sz.prod())
##### update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
pred_bboxes.append(pred_bbox)
# scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2],
pred_bbox[1] + pred_bbox[3]), (0, 255, 255),
3)
cv2.putText(img, str(idx), (40, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
if 'VOT2018-LT' == args.dataset:
video_path = os.path.join(save_dir, args.dataset, model_name,
'longterm', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
result_path = os.path.join(
video_path, '{}_001_confidence.value'.format(video.name))
with open(result_path, 'w') as f:
for x in scores:
f.write('\n') if x is None else f.write(
"{:.6f}\n".format(x))
result_path = os.path.join(video_path,
'{}_time.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in track_times:
f.write("{:.6f}\n".format(x))
elif 'GOT-10k' == args.dataset:
video_path = os.path.join(save_dir, args.dataset, model_name,
video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
result_path = os.path.join(video_path,
'{}_time.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in track_times:
f.write("{:.6f}\n".format(x))
else:
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path,
'{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.
format(v_idx + 1, video.name, toc, idx / toc))
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
model_name = args.tracker_name + '_' + args.tracker_param
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
if args.dataset in ['VOT2016', 'VOT2018', 'VOT2019']:
total_lost = 0
# restart tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
frame_counter = 0
lost_number = 0
toc = 0
'''对refinement module计时'''
toc_refine = 0
pred_bboxes = []
for idx, (img, gt_bbox) in enumerate(video):
if len(gt_bbox) == 4:
gt_bbox = [
gt_bbox[0], gt_bbox[1], gt_bbox[0],
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1,
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1, gt_bbox[1]
]
tic = cv2.getTickCount()
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
if idx == frame_counter:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(
gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
pred_bbox = gt_bbox_
pred_bboxes.append(1)
elif idx > frame_counter:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
overlap = vot_overlap(pred_bbox, gt_bbox,
(img.shape[1], img.shape[0]))
if overlap > 0:
# not lost
pred_bboxes.append(pred_bbox)
else:
# lost object
pred_bboxes.append(2)
frame_counter = idx + 5 # skip 5 frames
lost_number += 1
else:
pred_bboxes.append(0)
toc += cv2.getTickCount() - tic
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > frame_counter:
cv2.polylines(
img, [np.array(gt_bbox, np.int).reshape(
(-1, 1, 2))], True, (0, 255, 0), 3)
if len(pred_bbox) == 8:
cv2.polylines(
img,
[np.array(pred_bbox, np.int).reshape(
(-1, 1, 2))], True, (0, 255, 255), 3)
else:
bbox = list(map(int, pred_bbox))
cv2.rectangle(img, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.putText(img, str(lost_number), (40, 80),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
video_path = os.path.join(save_dir, args.dataset, model_name,
'baseline', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
if isinstance(x, int):
f.write("{:d}\n".format(x))
else:
f.write(','.join([vot_float2str("%.4f", i)
for i in x]) + '\n')
print(
'({:3d}) Video: {:12s} Time: {:4.1f}s Speed: {:3.1f}fps Lost: {:d}'
.format(v_idx + 1, video.name, toc, idx / toc, lost_number))
total_lost += lost_number
print("{:s} total lost: {:d}".format(model_name, total_lost))
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
model_name = args.tracker_name + '_' + args.tracker_param + '_%d' % (
args.run_id)
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
pred_bbox = gt_bbox_
scores.append(None)
if 'VOT2018-LT' == args.dataset:
pred_bboxes.append([1])
else:
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
'''##### refine tracking results #####'''
pred_bboxes.append(pred_bbox)
# scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main(model_code):
RF_module = RefineModule(refine_path.format(model_code),
selector_path,
search_factor=sr,
input_sz=input_sz)
model_name = 'AR_' + '{}'.format(
RF_type.format(model_code)) + '_iter_%d' % (n_iter)
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
if args.dataset in ['VOT2016', 'VOT2018', 'VOT2019']:
# restart tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
frame_counter = 0
lost_number = 0
toc = 0
pred_bboxes = []
for idx, (img, gt_bbox) in enumerate(video):
if len(gt_bbox) == 4:
gt_bbox = [
gt_bbox[0], gt_bbox[1], gt_bbox[0],
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1,
gt_bbox[1] + gt_bbox[3] - 1,
gt_bbox[0] + gt_bbox[2] - 1, gt_bbox[1]
]
tic = cv2.getTickCount()
if idx == frame_counter:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''##### initilize refinement module for specific video'''
RF_module.initialize(cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(gt_bbox_))
pred_bbox = gt_bbox_
pred_bboxes.append(1)
elif idx > frame_counter:
'''##### refine tracking results #####'''
for _ in range(n_iter):
pred_bbox = RF_module.refine(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h,
(H, W))
w = x2 - x1
h = y2 - y1
pred_bbox = np.array([x1, y1, w, h])
overlap = vot_overlap(pred_bbox, gt_bbox,
(img.shape[1], img.shape[0]))
if overlap > 0:
# not lost
pred_bboxes.append(pred_bbox)
else:
# lost object
pred_bboxes.append(2)
frame_counter = idx + 5 # skip 5 frames
lost_number += 1
else:
pred_bboxes.append(0)
toc += cv2.getTickCount() - tic
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > frame_counter:
cv2.polylines(
img, [np.array(gt_bbox, np.int).reshape(
(-1, 1, 2))], True, (0, 255, 0), 3)
bbox = list(map(int, pred_bbox))
cv2.rectangle(img, (bbox[0], bbox[1]),
(bbox[0] + bbox[2], bbox[1] + bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.putText(img, str(lost_number), (40, 80),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
video_path = os.path.join(save_dir, args.dataset, model_name,
'baseline', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
if isinstance(x, int):
f.write("{:d}\n".format(x))
else:
f.write(','.join([vot_float2str("%.4f", i)
for i in x]) + '\n')
print(
'({:3d}) Video: {:12s} Time: {:4.1f}s Speed: {:3.1f}fps Lost: {:d}'
.format(v_idx + 1, video.name, toc, idx / toc, lost_number))
def main():
RF_module = RefineModule(refine_path,
selector_path,
search_factor=sr,
input_sz=input_sz)
# refine_method = args.refine_method
model_name = 'siamrpn_' + RF_type + '_m2b_{}'.format(args.thres)
snapshot_path = os.path.join(
project_path_, 'experiments/%s/model.pth' % args.tracker_name)
config_path = os.path.join(
project_path_, 'experiments/%s/config.yaml' % args.tracker_name)
cfg.merge_from_file(config_path)
# create model
model = ModelBuilder() # a sub-class of `torch.nn.Module`
model = load_pretrain(model, snapshot_path).cuda().eval()
# build tracker
tracker = build_tracker(
model) # tracker is a object consisting of NN and some post-processing
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
tracker.init(img, gt_bbox_)
'''##### initilize refinement module for specific video'''
RF_module.initialize(cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
outputs = tracker.track(img)
pred_bbox = outputs['bbox']
'''##### refine tracking results #####'''
mask_pred = RF_module.get_mask(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB), np.array(pred_bbox))
pred_bbox = mask2bbox(mask_pred,
pred_bbox,
MASK_THRESHOLD=args.thres)
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
pred_bbox = np.array([x1, y1, w, h])
tracker.center_pos = np.array([x1 + w / 2, y1 + h / 2])
tracker.size = np.array([w, h])
pred_bboxes.append(pred_bbox)
scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset,
model_name + '_' + str(selector_path))
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
'''##### build a Refinement module #####'''
RF_module = RefineModule(refine_path,
selector_path,
search_factor=sr,
input_sz=input_sz)
model_name = 'RTMDNet' + '{}-{}'.format(RF_type, selector_path)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
'''build tracker'''
tracker = RT_MDNet()
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # RGB format
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''initialize tracker'''
tracker.initialize_seq(img_RGB, np.array(gt_bbox_))
'''initilize refine module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
if 'VOT2018-LT' == args.dataset:
pred_bboxes.append([1])
else:
pred_bboxes.append(pred_bbox)
else:
ori_bbox = tracker.track(img_RGB)
'''##### refine tracking results #####'''
pred_bbox = RF_module.refine(img_RGB, np.array(ori_bbox))
'''boundary and size limit'''
pred_bbox = bbox_clip(pred_bbox, (H, W))
'''update state'''
tracker.target_bbox = pred_bbox.copy()
pred_bboxes.append(pred_bbox)
# scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
ori_bbox = list(map(int, ori_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 0, 255), 3)
cv2.rectangle(
img, (ori_bbox[0], ori_bbox[1]),
(ori_bbox[0] + ori_bbox[2], ori_bbox[1] + ori_bbox[3]),
(255, 0, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 0), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
model_name = 'RTMDNet-oracle'
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
'''build tracker'''
tracker = RT_MDNet()
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
img_RGB = cv2.cvtColor(img, cv2.COLOR_BGR2RGB) # RGB format
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''initialize tracker'''
tracker.initialize_seq(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
ori_bbox = tracker.track(img_RGB)
pred_bbox = bbox_clip(ori_bbox, (H, W))
oracle_box = pred_bbox.copy()
cx, cy, _, _ = get_axis_aligned_bbox(np.array(gt_bbox))
oracle_box[:2] = np.array([cx, cy]) - oracle_box[2:] / 2
tracker.target_bbox = oracle_box
pred_bboxes.append(pred_bbox)
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
ori_bbox = list(map(int, ori_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 0, 255), 3)
cv2.rectangle(img, (oracle_box[0], oracle_box[1]),
(oracle_box[0] + oracle_box[2],
oracle_box[1] + oracle_box[3]), (255, 0, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 0), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
'''Refinement module'''
RF_module = RefineModule(refine_path,
selector_path,
search_factor=sr,
input_sz=input_sz)
model_name = args.tracker_name + '_' + args.tracker_param + '{}-{}'.format(
RF_type, selector_path) + '_%d' % (args.run_id)
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
color = np.array(COLORS[random.randint(0,
len(COLORS) - 1)])[None,
None, ::-1]
vis_result = os.path.join(
'/home/zxy/Desktop/AlphaRefine/CVPR21/material/quality_analysis/mask_vis',
'{}'.format(video.name))
if args.video != '':
# test one special video
if video.name != args.video:
continue
else:
print()
if not os.path.exists(vis_result):
os.makedirs(vis_result)
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
'''##### initilize refinement module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
'''##### refine tracking results #####'''
pred_bbox = RF_module.refine(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB), np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
new_pos = torch.from_numpy(
np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(
np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() /
tracker.base_target_sz.prod())
##### update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
mask_pred = RF_module.get_mask(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB), np.array(pred_bbox))
from external.pysot.toolkit.visualization import draw_mask
draw_mask(img,
mask_pred,
idx=idx,
show=True,
save_dir='dimpsuper_armask_crocodile-3')
pred_bboxes.append(pred_bbox)
# scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
im4show = img
mask_pred = np.uint8(mask_pred > 0.5)[:, :, None]
contours, _ = cv2.findContours(mask_pred.squeeze(),
cv2.RETR_LIST,
cv2.CHAIN_APPROX_SIMPLE)
im4show = im4show * (1 - mask_pred) + np.uint8(
im4show * mask_pred /
2) + mask_pred * np.uint8(color) * 128
pred_bbox = list(map(int, pred_bbox))
# gt_bbox = list(map(int, gt_bbox))
# cv2.rectangle(im4show, (gt_bbox[0], gt_bbox[1]),
# (gt_bbox[0]+gt_bbox[2], gt_bbox[1]+gt_bbox[3]), (0, 255, 0), 3)
# cv2.rectangle(im4show, (pred_bbox[0], pred_bbox[1]),
# (pred_bbox[0]+pred_bbox[2], pred_bbox[1]+pred_bbox[3]), color[::-1].squeeze().tolist(), 3)
cv2.drawContours(im4show, contours, -1, color[::-1].squeeze(),
2)
cv2.putText(im4show, str(idx), (40, 40),
cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
# cv2.imshow(video.name, im4show)
cv2.imwrite(os.path.join(vis_result, '{:06}.jpg'.format(idx)),
im4show)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
model_name = 'atom_oracle'
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
x1, y1, w, h = pred_bbox
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
pred_bbox = np.array([x1, y1, w, h])
'''##### reset tracking results #####'''
cx, cy, _, _ = get_axis_aligned_bbox(np.array(gt_bbox))
if not gt_bbox == [0, 0, 0, 0]:
new_pos = torch.from_numpy(
np.array([cy, cx]).astype(np.float32))
else:
new_pos = torch.from_numpy(
np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(
np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() /
tracker.base_target_sz.prod())
##### update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
pred_bboxes.append(pred_bbox)
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {'use_visdom': False, 'server': '127.0.0.1', 'port': 8097}
tracker = tracker_info.tracker_class(params)
'''Refinement module'''
RF_module = RefineModule(refine_path, selector_path, search_factor=sr, input_sz=input_sz)
model_name = args.tracker_name + '_' + args.tracker_param + '{}-{}'.format(RF_type, selector_path) + '_%d'%(args.run_id)
model_name = 'LaSOT_gt'
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset, dataset_root=dataset_root_, load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(os.path.join(save_dir, args.dataset, model_name, '{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
pred_bboxes = []
for idx, (img, gt_bbox) in enumerate(video):
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
pred_bboxes.append(gt_bbox_)
continue
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx-(w-1)/2, cy-(h-1)/2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
'''##### initilize refinement module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
''' refine tracking results '''
pred_bbox = RF_module.refine(cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
w, h = get_mean_wh(pred_bboxes, w, h)
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
new_pos = torch.from_numpy(np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() / tracker.base_target_sz.prod())
# update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
pred_bboxes.append(pred_bbox)
toc += cv2.getTickCount() - tic
track_times.append((cv2.getTickCount() - tic)/cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0]+gt_bbox[2], gt_bbox[1]+gt_bbox[3]), (0, 255, 0), 3)
cv2.rectangle(img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0]+pred_bbox[2], pred_bbox[1]+pred_bbox[3]), (0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
k = cv2.waitKey(0)
if k == ord('q'):
exit()
elif k == ord('s'):
cv2.imwrite(os.path.join(os.environ['HOME'], 'Desktop/demo', video.name+'_{}.jpg'.format(idx)), img)
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x])+'\n')
print(video.name)
def main():
model_name = 'siamRPN'
snapshot_path = os.path.join(
project_path_, 'experiments/%s/model.pth' % args.tracker_name)
config_path = os.path.join(
project_path_, 'experiments/%s/config.yaml' % args.tracker_name)
cfg.merge_from_file(config_path)
# create model
model = ModelBuilder() # a model is a Neural Network.(a torch.nn.Module)
# load model
model = load_pretrain(model, snapshot_path).cuda().eval()
# build tracker
tracker = build_tracker(
model
) # a tracker is a object consisting of not only a NN and some post-processing
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
tracker.init(img, gt_bbox_)
pred_bboxes.append(gt_bbox_)
else:
outputs = tracker.track(img)
pred_bbox = outputs['bbox']
x1, y1, w, h = pred_bbox
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
pred_bbox = np.array([x1, y1, w, h])
tracker.center_pos = np.array([x1 + w / 2, y1 + h / 2])
tracker.size = np.array([w, h])
pred_bboxes.append(pred_bbox)
scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# create tracker
tracker_info = Tracker(args.tracker_name, args.tracker_param, None)
params = tracker_info.get_parameters()
params.visualization = args.vis
params.debug = args.debug
params.visdom_info = {
'use_visdom': False,
'server': '127.0.0.1',
'port': 8097
}
tracker = tracker_info.tracker_class(params)
# setup refine module
RF_module = RefineModule(refine_path,
selector_path,
search_factor=sr,
input_sz=input_sz)
model_name = args.tracker_name + '_' + args.tracker_param + '{}-{}'.format(
RF_type, selector_path) + '_%d' % (args.run_id)
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root_,
load_img=False)
# OPE tracking
for v_idx, video in enumerate(dataset):
if os.path.exists(
os.path.join(save_dir, args.dataset, model_name,
'{}.txt'.format(video.name))):
continue
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
'''get RGB format image'''
img_RGB = img[:, :, ::-1].copy() # BGR --> RGB
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
'''Initialize'''
gt_bbox_np = np.array(gt_bbox_)
gt_bbox_torch = torch.from_numpy(gt_bbox_np.astype(np.float32))
init_info = {}
init_info['init_bbox'] = gt_bbox_torch
_ = tracker.initialize(img_RGB, init_info)
'''##### initilize refinement module for specific video'''
RF_module.initialize(img_RGB, np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
if 'VOT2018-LT' == args.dataset:
pred_bboxes.append([1])
else:
pred_bboxes.append(pred_bbox)
else:
'''Track'''
outputs = tracker.track(img_RGB)
pred_bbox = outputs['target_bbox']
'''##### refine tracking results #####'''
pred_bbox = RF_module.refine(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB), np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
new_pos = torch.from_numpy(
np.array([y1 + h / 2, x1 + w / 2]).astype(np.float32))
new_target_sz = torch.from_numpy(
np.array([h, w]).astype(np.float32))
new_scale = torch.sqrt(new_target_sz.prod() /
tracker.base_target_sz.prod())
##### update
tracker.pos = new_pos.clone()
tracker.target_sz = new_target_sz
tracker.target_scale = new_scale
pred_bboxes.append(pred_bbox)
# scores.append(outputs['best_score'])
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
if 'VOT2018-LT' == args.dataset:
video_path = os.path.join(save_dir, args.dataset, model_name,
'longterm', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
result_path = os.path.join(
video_path, '{}_001_confidence.value'.format(video.name))
with open(result_path, 'w') as f:
for x in scores:
f.write('\n') if x is None else f.write(
"{:.6f}\n".format(x))
result_path = os.path.join(video_path,
'{}_time.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in track_times:
f.write("{:.6f}\n".format(x))
elif 'GOT-10k' == args.dataset:
video_path = os.path.join(save_dir, args.dataset, model_name,
video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path,
'{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
result_path = os.path.join(video_path,
'{}_time.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in track_times:
f.write("{:.6f}\n".format(x))
else:
model_path = os.path.join(save_dir, args.dataset, model_name)
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
# refine_method = args.refine_method
n_iter = 2 # number of the the refine iterations
model_name = 'RF_' + RF_type + 'iter-{}'.format(n_iter)
dataset_root = dataset_root_
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset,
dataset_root=dataset_root,
load_img=False)
# create refine module
RF_module = RefineModule(refine_path,
selector_path,
search_factor=sr,
input_sz=input_sz)
# OPE tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
toc = 0
pred_bboxes = []
scores = []
track_times = []
for idx, (img, gt_bbox) in enumerate(video):
tic = cv2.getTickCount()
if idx == 0:
H, W, _ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx - (w - 1) / 2, cy - (h - 1) / 2, w, h]
# initialize refinement module for specific video
RF_module.initialize(cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(gt_bbox_))
pred_bbox = gt_bbox_
scores.append(None)
pred_bboxes.append(pred_bbox)
else:
# refine tracking results
for i in range(n_iter):
pred_bbox = RF_module.refine(
cv2.cvtColor(img, cv2.COLOR_BGR2RGB),
np.array(pred_bbox))
x1, y1, w, h = pred_bbox.tolist()
'''add boundary and min size limit'''
x1, y1, x2, y2 = bbox_clip(x1, y1, x1 + w, y1 + h, (H, W))
w = x2 - x1
h = y2 - y1
pred_bbox = np.array([x1, y1, w, h])
pred_bboxes.append(pred_bbox)
scores.append(None)
toc += cv2.getTickCount() - tic
track_times.append(
(cv2.getTickCount() - tic) / cv2.getTickFrequency())
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > 0:
gt_bbox = list(map(int, gt_bbox))
pred_bbox = list(map(int, pred_bbox))
cv2.rectangle(
img, (gt_bbox[0], gt_bbox[1]),
(gt_bbox[0] + gt_bbox[2], gt_bbox[1] + gt_bbox[3]),
(0, 255, 0), 3)
cv2.rectangle(
img, (pred_bbox[0], pred_bbox[1]),
(pred_bbox[0] + pred_bbox[2], pred_bbox[1] + pred_bbox[3]),
(0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX,
1, (0, 255, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
model_path = os.path.join(save_dir, args.dataset,
model_name + '_' + str(selector_path))
if not os.path.isdir(model_path):
os.makedirs(model_path)
result_path = os.path.join(model_path, '{}.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
f.write(','.join([str(i) for i in x]) + '\n')
print('({:3d}) Video: {:12s} Time: {:5.1f}s Speed: {:3.1f}fps'.format(
v_idx + 1, video.name, toc, idx / toc))
def main():
model_name = 'siamRPN'
snapshot_path = os.path.join(project_path_, 'experiments/%s/model.pth' % args.tracker_name)
config_path = os.path.join(project_path_, 'experiments/%s/config.yaml' % args.tracker_name)
cfg.merge_from_file(config_path)
# create model
model = ModelBuilder() # a model is a Neural Network.(a torch.nn.Module)
# load model
model = load_pretrain(model, snapshot_path).cuda().eval()
# build tracker
tracker = build_tracker(model) # a tracker is a object consisting of not only a NN and some post-processing
# create dataset
dataset = DatasetFactory.create_dataset(name=args.dataset, dataset_root=dataset_root_, load_img=False)
total_lost = 0
if args.dataset in ['VOT2016', 'VOT2018', 'VOT2019']:
# restart tracking
for v_idx, video in enumerate(dataset):
if args.video != '':
# test one special video
if video.name != args.video:
continue
frame_counter = 0
lost_number = 0
toc = 0
pred_bboxes = []
for idx, (img, gt_bbox) in enumerate(video):
if len(gt_bbox) == 4:
gt_bbox = [gt_bbox[0], gt_bbox[1],
gt_bbox[0], gt_bbox[1]+gt_bbox[3]-1,
gt_bbox[0]+gt_bbox[2]-1, gt_bbox[1]+gt_bbox[3]-1,
gt_bbox[0]+gt_bbox[2]-1, gt_bbox[1]]
tic = cv2.getTickCount()
if idx == frame_counter:
H,W,_ = img.shape
cx, cy, w, h = get_axis_aligned_bbox(np.array(gt_bbox))
gt_bbox_ = [cx-(w-1)/2, cy-(h-1)/2, w, h]
tracker.init(img, gt_bbox_)
pred_bbox = gt_bbox_
pred_bboxes.append(1)
elif idx > frame_counter:
outputs = tracker.track(img)
pred_bbox = outputs['bbox']
overlap = vot_overlap(pred_bbox, gt_bbox, (img.shape[1], img.shape[0]))
if overlap > 0:
# not lost
pred_bboxes.append(pred_bbox)
else:
# lost object
pred_bboxes.append(2)
frame_counter = idx + 5 # skip 5 frames
lost_number += 1
else:
pred_bboxes.append(0)
toc += cv2.getTickCount() - tic
if idx == 0:
cv2.destroyAllWindows()
if args.vis and idx > frame_counter:
cv2.polylines(img, [np.array(gt_bbox, np.int).reshape((-1, 1, 2))],
True, (0, 255, 0), 3)
bbox = list(map(int, pred_bbox))
cv2.rectangle(img, (bbox[0], bbox[1]),
(bbox[0]+bbox[2], bbox[1]+bbox[3]), (0, 255, 255), 3)
cv2.putText(img, str(idx), (40, 40), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 255, 255), 2)
cv2.putText(img, str(lost_number), (40, 80), cv2.FONT_HERSHEY_SIMPLEX, 1, (0, 0, 255), 2)
cv2.imshow(video.name, img)
cv2.waitKey(1)
toc /= cv2.getTickFrequency()
# save results
video_path = os.path.join(save_dir, args.dataset, model_name,
'baseline', video.name)
if not os.path.isdir(video_path):
os.makedirs(video_path)
result_path = os.path.join(video_path, '{}_001.txt'.format(video.name))
with open(result_path, 'w') as f:
for x in pred_bboxes:
if isinstance(x, int):
f.write("{:d}\n".format(x))
else:
f.write(','.join([vot_float2str("%.4f", i) for i in x])+'\n')
print('({:3d}) Video: {:12s} Time: {:4.1f}s Speed: {:3.1f}fps Lost: {:d}'.format(
v_idx+1, video.name, toc, idx / toc, lost_number))