def background_flow(next_image, current_bbox, backward_flow):
mask = np.zeros(next_image.shape) + 0.3
mask[current_bbox[1]:current_bbox[3],
current_bbox[0]:current_bbox[2], :] = 1
next_mask = flow_warp_backward(mask, backward_flow)
return next_mask
def LAB_mask(im1,im2,flow):
warped_im1 = flow_warp_backward(im1, flow)
diff, rgb_diff = color_confidence(warped_im1, im2)
diff = diff/(np.max(diff)-np.min(diff))
diff_mask = np.zeros((np.shape(diff)[0],np.shape(diff)[1],3),dtype='float')
diff_mask[:,:,0] = diff
diff_mask[:,:,1] = diff
diff_mask[:,:,2] = diff
#np.save(os.path.join(data_dir,format(i+1, '08')+"_lab.npy"), diff_mask)
return diff_mask
next_bbox = bbox_format(gts[i + 1], 'tlxy_wh_2_rect')
next_bbox = [int(j) for j in next_bbox]
flow = np.load(
os.path.join(confidence_dir,
format(i + 1, '08') + '_flow.npy'))
entropy_data = np.load(
os.path.join(confidence_dir,
format(i + 1, '08') + '_entropy.npy'))
entropy = (entropy_data - np.min(entropy_data)) / (
np.max(entropy_data) - np.min(entropy_data))
confidence = 1 - entropy
warped_im1 = flow_warp(im1, flow)
diff, rgb_diff = color_confidence(warped_im1, im2)
#next_mask = background_flow(im2, detection_box, flow)
next_mask = flow_warp_backward(foreground_candidate_mask,
flow)
# np.save(os.path.join(data_dir,format(i+1, '08')+"_fgmask.npy"), next_mask)
if (i > 0):
pre_flow_mask = np.load(
os.path.join(data_dir,
format(i, '08') + "_fgmask.npy"))
double_warped = flow_warp(pre_flow_mask, flow)
# np.save(os.path.join(data_dir,format(i+1, '08')+"_prev_fgmask.npy"), double_warped)
overall_mask = (confidence[:, :, 0] * next_mask[:, :, 0] +
entropy[:, :, 0]) / 2
state = SiamRPN_track(state, im2, overall_mask,
confidence[:, :, 0],
next_mask[:, :, 0], entropy[:, :, 0],
index_1, index_2, gts[i + 1])
def generate_masks(index, train_im_paths, train_labels):
im_paths = train_im_paths[index]
labels = train_labels[index]
flow_data = []
entropy_data = []
scale_data = []
lab_data = []
iou_data = []
detection_data = []
box_data = []
warped_images = []
box_data = []
for i in range(0, 3):
im = np.copy(cv2.imread(im_paths[i]))
im_h, im_w, im_c = im.shape
names = im_paths[i].split('/')[-1].split('_')
if (len(names) > 4):
video_name = ''
for more_index in range(0, len(names) - 3):
if (more_index > 0):
video_name += '_'
video_name += names[more_index]
frame_name = names[len(names) - 3]
type_name = names[len(names) - 2]
else:
video_name = names[0]
frame_name = names[1]
type_name = names[2]
class_dir = os.path.join(data_dir, type_name, video_name)
#x0,y0,w,h
gt = youtube_to_rec(labels[i], im_h, im_w)
flow = np.zeros((im_h, im_w, 2))
if (i == 0):
filename = os.path.join(class_dir, str(frame_name))
with open(filename + '.txt') as ff:
content = ff.readlines()
ff.close()
iou_result = []
detection_score = []
temp_result = []
for j in range(0, 100):
data = content[100 * i + j].split()
x0, y0, w, h = data[0:4]
res = [int(x0), int(y0), int(w), int(h)]
temp_result.append(res)
iou_result.append(float(data[4]))
detection_score.append(float(data[5]))
iou_data.append(iou_result)
detection_data.append(detection_score)
box_data.append(temp_result)
if (i > 0):
#im = cv2.imread(im_paths[i-1])
flow = np.load(
os.path.join(class_dir,
str(frame_name) + '_flow.npy'))
entropy = np.load(
os.path.join(class_dir,
str(frame_name) + '_entropy.npy'))
warped_im = flow_warp_backward(im, flow)
diff, rgb_diff = color_confidence(warped_im, im)
flow_data.append(flow)
lab_data.append(diff)
entropy_data.append(entropy)
return flow_data, entropy_data, lab_data, iou_data, detection_data, box_data
