def show_boxes_from_standard_json(json_file_path,
classes,
img_folder_path=None,
output_folder_path=None,
track_id=-1):
dets = read_json_from_file(json_file_path)
for det in dets:
python_data = det
if img_folder_path is None:
img_path = os.path.join(python_data["image"]["folder"],
python_data["image"]["name"])
else:
img_path = os.path.join(img_folder_path,
python_data["image"]["name"])
if is_image(img_path): img = cv2.imread(img_path)
candidates = python_data["candidates"]
for candidate in candidates:
bbox = np.array(candidate["det_bbox"]).astype(int)
score = candidate["det_score"]
if score >= bbox_thresh:
img = draw_bbox(img, bbox, score, classes, track_id=track_id)
if output_folder_path is not None:
create_folder(output_folder_path)
img_output_path = os.path.join(output_folder_path,
python_data["image"]["name"])
cv2.imwrite(img_output_path, img)
return True
def show_poses_from_standard_json(json_file_path,
joint_pairs,
joint_names,
img_folder_path=None,
output_folder_path=None):
poses = read_json_from_file(json_file_path)
for pose in poses:
python_data = pose
if img_folder_path is None:
img_path = os.path.join(python_data["image"]["folder"],
python_data["image"]["name"])
else:
img_path = os.path.join(img_folder_path,
python_data["image"]["name"])
if is_image(img_path): img = cv2.imread(img_path)
candidates = python_data["candidates"]
for candidate in candidates:
pose_keypoints_2d = candidate["pose_keypoints_2d"]
joints = reshape_keypoints_into_joints(pose_keypoints_2d)
img = show_poses_from_python_data(img, joints, joint_pairs,
joint_names)
if output_folder_path is not None:
create_folder(output_folder_path)
img_output_path = os.path.join(output_folder_path,
python_data["image"]["name"])
cv2.imwrite(img_output_path, img)
return
def batch_test_images(folder_path_in, folder_path_out):
joint_names = [
'head', 'upper neck', 'right shoulder', 'right elbow', 'right wrist',
'left shoulder', 'left elbow', 'left wrist', 'right pelvis',
'right knee', 'right ankle', 'left pelvis', 'left knee', 'left ankle',
'background'
]
joint_pairs = [['head', 'upper neck', 'purple'],
['upper neck', 'right shoulder', 'yellow'],
['upper neck', 'left shoulder', 'yellow'],
['right shoulder', 'right elbow', 'blue'],
['right elbow', 'right wrist', 'green'],
['left shoulder', 'left elbow', 'blue'],
['left elbow', 'left wrist', 'green'],
['right shoulder', 'right pelvis', 'yellow'],
['left shoulder', 'left pelvis', 'yellow'],
['right pelvis', 'right knee', 'red'],
['right knee', 'right ankle', 'skyblue'],
['left pelvis', 'left knee', 'red'],
['left knee', 'left ankle', 'skyblue']]
# setup paths
deployFile = deploy_proto_path
caffemodel = caffe_model_prefix + '.caffemodel'
norm_size = img_size
print('deployFile = %s\n', deployFile)
print('caffemodel = %s\n', caffemodel)
# load network
if flag_GPU is True:
caffe.set_mode_gpu()
caffe.set_device(flag_GPU_id)
else:
caffe.set_mode_cpu()
net = caffe.Net(deployFile, caffemodel, caffe.TEST)
print('testing scales: ', scales)
print('output_image_folder_path: ', output_image_folder_path)
# Do the batch test
subfolder_names = get_immediate_subfolder_names(folder_path_in)
for subfolder_name in subfolder_names:
subfolder_path_in = os.path.join(folder_path_in, subfolder_name)
img_names = get_immediate_childfile_names(subfolder_path_in)
for ct, img_name in enumerate(img_names):
if flag_selective is True and img_name not in IMG_NAMES_TO_SHOW:
continue
print("processing: ", img_name, ct, len(img_names))
# load image
img_path_in = os.path.join(subfolder_path_in, img_name)
if is_image(img_path_in): img_raw = cv2.imread(img_path_in)
# find heatmaps
# OPTIONAL: flip and zoom
if flag_flip is True and flag_zoom is True:
heatmaps_from_multi_res = process_img_scales_and_flips(
net, img_raw, norm_size, scales, heatmap_layer_name)
# OPTIONAL: flip
elif flag_flip is True:
heatmaps_from_multi_res = [
process_img_flip(net, img_raw, norm_size),
img_to_heatmaps(net, img_raw, norm_size,
heatmap_layer_name)
]
# OPTIONAL: zoom
elif flag_zoom is True:
heatmaps_from_multi_res = process_img_scales(
net, img_raw, norm_size, scales)
else:
heatmaps = img_to_heatmaps(net, img_raw, norm_size,
heatmap_layer_name)
# Average heatmaps
if flag_flip is True or flag_zoom is True:
heatmaps = average_multiple_heatmap_sets(
heatmaps_from_multi_res)
# Find joints from heatmaps
if flag_nms is True:
joints = find_joints_in_heatmaps_nms(heatmaps)
else:
joints = find_joints_in_heatmaps(heatmaps)
# Draw joints on image
img_demo = cv2.resize(img_raw, (norm_size, norm_size))
img_demo = demo_poses_in_img(img_demo, joints, joint_pairs,
joint_names)
# Save images to a path
if flag_save_images is True:
create_folder(os.path.join(folder_path_out, subfolder_name))
img_name_out = os.path.splitext(img_name)[0] + '.png'
img_path_out = os.path.join(folder_path_out, subfolder_name,
img_name_out)
cv2.imwrite(img_path_out, img_demo)
# Demo Heatmaps
if flag_demo_heatmaps is True:
demo_heatmaps(heatmaps, joint_names)
# Save joint predictions for MPII dataset evaluation
if flag_save_evaluation is True:
save_folder = os.path.join(output_image_folder_path,
'quantitative')
create_folder(save_folder)
if test_dataset == 'test_MPII' or test_dataset == 'validate_MPII':
rect_id = find_rect_id(img_name)
save_pose_preditions(joints, save_folder, img_name,
rect_id)
elif test_dataset == 'test_LSP':
save_pose_preditions(joints, save_folder, img_name)
def draw_shelf_bbox(img, joints, joint_pairs, joint_names, track_id = -1, trigger = False, img_id = -1, img_path=None, pick_results=[]):
shelf_bbox_pixels = [[[[706, 342],[707, 313],[823, 463],[817, 494]],
[[709, 308],[713, 275],[833, 416],[825, 454]],
[[713, 266],[716, 226],[849, 365],[838, 408]],
[[719, 221],[722, 172],[868, 297],[853, 357]],
[[723, 170],[728, 107],[888, 214],[875, 288]],
[[729, 104],[738, 20],[934, 110],[887, 213]]],
[[[ 818, 496],[ 827, 459],[ 987, 661],[ 964, 698]],
[[ 831, 450],[ 839, 415],[1015, 614],[ 994, 650]],
[[ 846, 407],[ 851, 368],[1044, 560],[1023, 606]],
[[ 858, 356],[ 869, 299],[1088, 484],[1052, 551]],
[[ 877, 290],[ 889, 231],[1141, 405],[1098, 474]],
[[ 896, 221],[ 924, 115],[1248, 263],[1154, 392]]]]
color = (0, 255, 0) #green
font = cv2.FONT_HERSHEY_SIMPLEX
for i, shelf in enumerate(shelf_bbox_pixels):
for j, pixels in enumerate(shelf):
pts = np.array(pixels,np.int32)
pts = pts.reshape((-1,1,2))
cv2.polylines(img,[pts],isClosed = True,color = color,thickness = 2)
cv2.putText(img,
#'{:s} {:.2f}'.format("ID:"+str(track_id), score),
'{:s}'.format(str(i+1)+"-"+str(j+1)),
(pts[0][0][0]-10, pts[0][0][1]-5),
font,
fontScale=0.8,
color=color,
thickness = 2,
lineType = cv2.LINE_AA)
if trigger:
ind_1 = joint_names.index('right elbow')
ind_2 = joint_names.index('right wrist')
x1, y1, sure1 = joints[ind_1]
x2, y2, sure2 = joints[ind_2]
x3 = int(x2 + (x2-x1) / 4)
y3 = int(y2 + (y2-y1) / 4)
cv2.circle(img, (x3, y3), radius=3, color=(255,255,255), thickness=2)
for i, shelf in enumerate(shelf_bbox_pixels):
for j, pixels in enumerate(shelf):
if isPoiWithinPoly([x3,y3], [pixels]):
pts = np.array(pixels,np.int32)
pts = pts.reshape((-1,1,2))
cv2.polylines(img,[pts],isClosed = True,color = (0,0,255),thickness = 2)
memory_img_id = img_id-50 #img of 50 frame before
memory_img_path = img_path[:-7] + str(memory_img_id) + img_path[-4:]
if is_image(memory_img_path):
memory_img = cv2.imread(memory_img_path)
cropImg = memory_img[y3-30:y3+30,x3-15:x3+45]
else:
print("No memory img {}.".format(memory_img_id))
pick_results.append({'track_id':track_id, 'shelf_id':[i+1,j+1], 'img_id':img_id, 'hand_pos':[x3,y3], 'memory_item_img':cropImg})
print("ID:"+str(track_id)+" picks on shelf "+ str(i+1)+ "-" + str(j+1) +" at time "+str(img_id))
break
vis_result_x = 30
vis_result_y = 30
for i, result in enumerate(pick_results):
cv2.putText(img,
'{:s}'.format("ID:"+str(result['track_id'])+" picks on shelf "+ str(result['shelf_id'][0])+ "-" + str(result['shelf_id'][1]) +" at time "+str(result['img_id'])),
(vis_result_x, vis_result_y),
font,
fontScale=0.8,
color=(255,255,255),
thickness = 2,
lineType = cv2.LINE_AA)
vis_result_y += 70
img[vis_result_y-60:vis_result_y, vis_result_x:vis_result_x+60] = result['memory_item_img']
vis_result_y += 30
return img, pick_results