def _test_traindata_provider():
seqs = load_seq_infos(1)
seq = seqs[0]
show_fid = TestCfg.SHOW_TRACK_RESULT_FID
trk = tracker.ConvRegTracker()
init = seq.gtRect[0]
init_rect = Rect(*init)
img_root = os.path.join(TestCfg.SEQUENCE_DIR, '../', seq.path)
path = os.path.join(img_root,
seq.imgFormat.format(seq.startFrame))
init_image = cv2.imread(path)
display.show_track_res(seq.startFrame, init_image, init_rect, init_rect, show_fid)
trk.init(init_image, init_rect)
while True:
frame_id = seq.startFrame
path = os.path.join(img_root,
seq.imgFormat.format(frame_id))
image = cv2.imread(path)
gt_rect = Rect(*seq.gtRect[0])
pred_rect = trk.track(image)
display.show_track_res(frame_id, image, gt_rect, pred_rect, show_fid)
def show_res(sec_image, obj_image, conf, out_conf, preprocess):
conf = conf.copy()
out_conf = out_conf.copy()
si = sec_image.copy()
si = np.asarray((si + 0.5) * 255.0, dtype=np.uint8)
si_w = si.shape[1]
si_h = si.shape[0]
ti = obj_image.copy()
ti = np.asarray((ti + 0.5) * 255.0, dtype=np.uint8)
conf_y, conf_x = np.unravel_index(np.argmax(out_conf), out_conf.shape)
predict_w = si_w / 2.5
predict_h = si_h / 2.5
srect = Rect(0, 0, sec_image.shape[1], sec_image.shape[0])
pcx, pcy = preprocess.predict_location(srect, conf_x, conf_y)
tlx = int(pcx - (predict_w - 1) / 2.0 + 0.5)
tly = int(pcy - (predict_h - 1) / 2.0 + 0.5)
obj_rect = Rect(tlx, tly, predict_w, predict_h).get_int_rect()
cv2.rectangle(si, obj_rect.get_tl(), obj_rect.get_dr(), (255, 0, 0), 2)
plt.figure(23245)
plt.clf()
plt.imshow(si)
plt.figure(12897)
plt.clf()
plt.subplot(131)
plt.imshow(ti)
plt.subplot(132)
plt.imshow(conf)
plt.colorbar()
plt.subplot(133)
plt.imshow(out_conf)
plt.colorbar()
plt.show()
plt.pause(0.1)
def _test_statistic_motion():
seqs = load_seq_infos()
data_list = []
img = np.zeros((2001,2001,3), dtype=np.uint8)
for seq in seqs:
_tmp = []
_rect = Rect(*seq.gtRect[0])
for i in range(1, len(seq.gtRect)):
_next_rect = Rect(*seq.gtRect[i])
_size = math.sqrt(_rect.w*_rect.h)
rel_motion_x = (_next_rect.get_center()[0] - _rect.get_center()[0])/_size
rel_motion_y = (_next_rect.get_center()[1] - _rect.get_center()[1])/_size
_tmp.append((rel_motion_x, rel_motion_y))
_rect = _next_rect
idx = round(rel_motion_x * 1000) + 1000
idy = round(rel_motion_y * 1000) + 1000
if 0 <= idx < 2001 and 0 <= idy < 2001:
img[idy, idx, 1] = 255
data_list.extend(_tmp)
print('Total length: {}'.format(len(data_list)))
data = np.array(data_list, dtype=np.float)
import scipy.stats
rv = scipy.stats.norm.fit(data[:], floc=0.0)
print(rv)
cv2.circle(img, (1000,1000), radius=int(round(1000*rv[1])), color=(255,0,0), thickness=3)
plt.figure()
plt.imshow(img)
plt.show()
print('Variance: {:.4f}'.format(rv[1]))
plt.waitforbuttonpress()
def _test_tracker():
seqs = load_seq_infos()
seqs.sort(key=lambda o: o.name)
show_fid = TestCfg.SHOW_TRACK_RESULT_FID
trk = tracker.ConvRegTracker()
seq = seqs[0]
init = seq.gtRect[0]
init_rect = Rect(*init)
img_root = os.path.join(TestCfg.SEQUENCE_DIR, '../', seq.path)
path = os.path.join(img_root,
seq.imgFormat.format(seq.startFrame))
init_image = cv2.imread(path)
display.show_track_res(seq.startFrame, init_image, init_rect, init_rect, show_fid)
trk.init(init_image, init_rect)
for fid in range(1, len(seq.gtRect)):
frame_id = fid + seq.startFrame
path = os.path.join(img_root,
seq.imgFormat.format(frame_id))
image = cv2.imread(path)
gt_rect = Rect(*seq.gtRect[fid])
pred_rect = trk.track(image)
display.show_track_res(frame_id, image, gt_rect, pred_rect, show_fid)
def run_tracker(s_frames, init_rect):
trker = ConvRegTracker()
image = cv2.imread(s_frames[0])
rect = Rect(*init_rect)
trker.init(image, rect)
res = []
res.append(list(init_rect))
for i in range(1, len(s_frames)):
image = cv2.imread(s_frames[i])
rect = trker.track(image)
res.append([rect.x, rect.y, rect.w, rect.h])
return res
def clip_image(image,rect):
iw = image.shape[1]
ih = image.shape[0]
im_rect = Rect(0,0,iw,ih)
if rect.is_in_rect(im_rect):
return image[rect.y:rect.y+rect.h,rect.x:rect.x+rect.w,:].copy()
xa = np.arange(rect.w)+rect.x
xa[xa<0] = 0
xa[xa>=iw] = iw-1
xa = np.tile(xa[None,:],(rect.h,1))
ya = np.arange(rect.h)+rect.y
ya[ya<0] = 0
ya[ya>=ih] = ih-1
ya = np.tile(ya[:,None],(1,rect.w))
return image[ya,xa]
def get_object_rect_by_index(self, search_rect, obj_index_y, obj_index_x):
_yi, _xi = obj_index_y, obj_index_x
_x_resolution = search_rect.w / float(self.feature_size_w)
_y_resolution = search_rect.h / float(self.feature_size_h)
_dyi = _yi - int((self.response_size_h-1) / 2.0)
_dxi = _xi - int((self.response_size_w-1) / 2.0)
patch_cx, patch_cy = search_rect.get_center()
pd_cx, pd_cy = patch_cx + _dxi*_x_resolution, patch_cy + _dyi*_y_resolution
_search_ratio_w = self.feature_size_w / float(self.convolution_w)
_search_ratio_h = self.feature_size_h / float(self.convolution_h)
pd_w, pd_h = round(search_rect.w/_search_ratio_w), round(search_rect.h/_search_ratio_h)
pd_tlx = round(pd_cx - (pd_w-1)/2.0)
pd_tly = round(pd_cy - (pd_h-1)/2.0)
final_rect = Rect(pd_tlx, pd_tly, pd_w, pd_h)
return final_rect
def get_scaled_search_feature(self, image, object_rect):
_scale_step_w = max(1, round(object_rect.w * self.scale_ratio))
_scale_step_h = max(1, round(object_rect.h * self.scale_ratio))
scaled_object_rects = []
for i in range(2 * self.scale_test_num + 1):
w = object_rect.w + _scale_step_w * (i - self.scale_test_num)
h = object_rect.h + _scale_step_h * (i - self.scale_test_num)
if w < 5 or h < 5:
print('Warning: w < 5 or h < 5')
continue
cx, cy = object_rect.get_center()
tl_x = round(cx - (w - 1)/2.0)
tl_y = round(cy - (h - 1)/2.0)
_rect = Rect(tl_x, tl_y, w, h)
scaled_object_rects.append(_rect)
_search_ratio_w = self.feature_size_w / float(self.convolution_w)
_search_ratio_h = self.feature_size_h / float(self.convolution_h)
_search_rect_list = []
_search_bgr_list = []
_search_input_list = []
for _scaled_rect in scaled_object_rects:
_search_rect = _scaled_rect.get_copy().scale_from_center(_search_ratio_w,
_search_ratio_h)
_search_bgr = clip_image(image, _search_rect)
_search_input = cv2.resize(_search_bgr, (self.input_search_w, self.input_search_h))
_search_rect_list.append(_search_rect)
_search_bgr_list.append(_search_bgr)
_search_input_list.append(_search_input)
if self._show_search_bgr_fid:
display.show_image(_search_bgr_list[0], self._show_search_bgr_fid, 'Train & search patch')
_search_features = self.extractor.extract_multiple_features(_search_input_list)
return _search_rect_list, _search_bgr_list, _search_features, scaled_object_rects
def _test_data_provider():
patch_rect = Rect(0,0, 500, 500)
gt_rect = Rect(152, 134, 42, 120)
response_size = (32, 32)