def __init__(self,
image_folder,
frames,
bboxes=None,
joints2d=None,
scale=1.0,
crop_size=224):
self.image_file_names = [
osp.join(image_folder, x) for x in os.listdir(image_folder)
if x.endswith('.png') or x.endswith('.jpg')
]
self.image_file_names = sorted(self.image_file_names)
self.image_file_names = np.array(self.image_file_names)[frames]
self.bboxes = bboxes
self.joints2d = joints2d
self.scale = scale
self.crop_size = crop_size
self.frames = frames
self.has_keypoints = True if joints2d is not None else False
self.norm_joints2d = np.zeros_like(self.joints2d)
if self.has_keypoints:
bboxes, time_pt1, time_pt2 = get_all_bbox_params(joints2d,
vis_thresh=0.3)
bboxes[:, 2:] = 150. / bboxes[:, 2:]
self.bboxes = np.stack(
[bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
self.image_file_names = self.image_file_names[time_pt1:time_pt2]
self.joints2d = joints2d[time_pt1:time_pt2]
self.frames = frames[time_pt1:time_pt2]
def __init__(self,
images,
frames,
bboxes=None,
joints2d=None,
scale=1.0,
crop_size=224):
self.images = images
self.bboxes = bboxes
self.joints2d = joints2d
self.scale = scale
self.crop_size = crop_size
self.frames = frames
self.has_keypoints = True if joints2d is not None else False
self.norm_joints2d = np.zeros_like(self.joints2d)
if self.has_keypoints:
bboxes, time_pt1, time_pt2 = get_all_bbox_params(joints2d,
vis_thresh=0.3)
bboxes[:, 2:] = 150. / bboxes[:, 2:]
self.bboxes = np.stack(
[bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
self.images = self.images[time_pt1:time_pt2]
self.joints2d = joints2d[time_pt1:time_pt2]
self.frames = frames[time_pt1:time_pt2]
def preprocess_video(video,
joints2d,
bboxes,
frames,
scale=1.0,
crop_size=224):
"""
Read video, do normalize and crop it according to the bounding box.
If there are bounding box annotations, use them to crop the image.
If no bounding box is specified but openpose detections are available, use them to get the bounding box.
:param video (ndarray): input video
:param joints2d (ndarray, NxJx3): openpose detections
:param bboxes (ndarray, Nx5): bbox detections
:param scale (float): bbox crop scaling factor
:param crop_size (int): crop width and height
:return: cropped video, cropped and normalized video, modified bboxes, modified joints2d
"""
if joints2d is not None:
bboxes, time_pt1, time_pt2 = get_all_bbox_params(joints2d,
vis_thresh=0.3)
bboxes[:, 2:] = 150. / bboxes[:, 2:]
bboxes = np.stack(
[bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
video = video[time_pt1:time_pt2]
joints2d = joints2d[time_pt1:time_pt2]
frames = frames[time_pt1:time_pt2]
shape = video.shape
temp_video = np.zeros((shape[0], crop_size, crop_size, shape[-1]))
norm_video = torch.zeros(shape[0], shape[-1], crop_size, crop_size)
for idx in range(video.shape[0]):
img = video[idx]
bbox = bboxes[idx]
j2d = joints2d[idx] if joints2d is not None else None
norm_img, raw_img, kp_2d = get_single_image_crop_demo(
img, bbox, kp_2d=j2d, scale=scale, crop_size=crop_size)
if joints2d is not None:
joints2d[idx] = kp_2d
temp_video[idx] = raw_img
norm_video[idx] = norm_img
temp_video = temp_video.astype(np.uint8)
return temp_video, norm_video, bboxes, joints2d, frames
def __init__(self,
image_folder,
bboxes=None,
joints2d=None,
scale=1.0,
crop_size=224):
self.image_file_names = [
osp.join(image_folder, x) for x in os.listdir(image_folder)
if x.endswith('.png') or x.endswith('.jpg')
]
self.image_file_names = sorted(self.image_file_names)
frames = [x for x in range(len(self.image_file_names))]
self.image_file_names = np.array(self.image_file_names)[frames]
print('#######################', self.image_file_names,
'#########################')
self.bboxes = bboxes
self.scale = scale
self.crop_size = crop_size
self.frames = frames
self.has_keypoints = True #if joints2d is not None else False
#self.norm_joints2d = np.zeros_like(self.joints2d)
self.annotation_path = image_folder + '/fitted.npy'
self.data = np.load(self.annotation_path,
encoding='latin1',
allow_pickle=True).item()
self.joints2d = [
self.data[textED(key)]['p2d'] for key in self.image_file_names
]
self.pose = np.asarray([
self.data[textED(key)]['fitting_params']['pose']
for key in self.image_file_names
])
# 3d keypoints
self.skeleton_smpl = np.asarray([
self.data[textED(key)]['fitting_params']['v']
for key in self.image_file_names
])
self.cam_t = np.asarray([
self.data[textED(key)]['fitting_params']['cam_t']
for key in self.image_file_names
])
#print('&&&&&&&&&&&3D SHAPE------',self.pose.shape)
joints2d = np.asarray(self.joints2d)
ones = np.ones((joints2d.shape[0], joints2d.shape[1], 3))
ones[:, :, :2] = joints2d
self.joints2d = ones #np.concatenate( (self.joints2d, np.ones( (len(self.joints2d),1) ) ), axis=1)
#print('2d GT shape-----------',self.joints2d.shape)
#print('2d GT -------------',joints2d)
if self.has_keypoints:
bboxes, time_pt1, time_pt2 = get_all_bbox_params(self.joints2d,
vis_thresh=0.3)
#print('box shape',bboxes)
bboxes[:, 2:] = 150. / bboxes[:, 2:]
self.bboxes = np.stack(
[bboxes[:, 0], bboxes[:, 1], bboxes[:, 2], bboxes[:, 2]]).T
#print('bboxes---',self.bboxes)
self.image_file_names = self.image_file_names[time_pt1:time_pt2]
self.joints2d = joints2d[time_pt1:time_pt2]
self.frames = frames[time_pt1:time_pt2]
