def get_fitting_error(self, coco_joint, smpl_mesh, cam_param, img2bb_trans,
coco_joint_valid):
# get coco joint from smpl mesh
coco_from_smpl = np.dot(self.coco_joint_regressor, smpl_mesh)
coco_from_smpl = self.add_pelvis(
coco_from_smpl) # z-axis component will be removed
coco_from_smpl = cam2pixel(coco_from_smpl, cam_param['focal'],
cam_param['princpt'])
coco_from_smpl_xy1 = np.concatenate(
(coco_from_smpl[:, :2], np.ones_like(coco_from_smpl[:, 0:1])), 1)
coco_from_smpl[:, :2] = np.dot(img2bb_trans,
coco_from_smpl_xy1.transpose(
1, 0)).transpose(1, 0)
coco_from_smpl[:, 0] = coco_from_smpl[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
coco_from_smpl[:, 1] = coco_from_smpl[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
# mask joint coordinates
coco_joint = coco_joint[:, :2][np.tile(coco_joint_valid, (
1, 2)) == 1].reshape(-1, 2)
coco_from_smpl = coco_from_smpl[:, :2][np.tile(coco_joint_valid, (
1, 2)) == 1].reshape(-1, 2)
error = np.sqrt(np.sum((coco_joint - coco_from_smpl)**2, 1)).mean()
return error
def get_bbox(joint_world, joint_valid, camrot, campos, focal, princpt):
joint_cam = []
for i in range(len(joint_world)):
joint_cam.append(world2cam(joint_world[i], camrot, campos))
joint_cam = np.array(joint_cam).reshape(-1, 3)
x_img, y_img, z_img = cam2pixel(joint_cam, focal, princpt)
x_img = x_img[joint_valid[:, 0] == 1]
y_img = y_img[joint_valid[:, 0] == 1]
xmin = min(x_img)
ymin = min(y_img)
xmax = max(x_img)
ymax = max(y_img)
x_center = (xmin + xmax) / 2.
width = xmax - xmin
xmin = x_center - 0.5 * width * 1.5
xmax = x_center + 0.5 * width * 1.5
y_center = (ymin + ymax) / 2.
height = ymax - ymin
ymin = y_center - 0.5 * height * 1.5
ymax = y_center + 0.5 * height * 1.5
bbox = np.array([xmin, ymin, xmax, ymax]).astype(np.float32)
return bbox
def __init__(self, transform, mode, annot_subset):
self.mode = mode # train, test, val
self.annot_subset = annot_subset # all, human_annot, machine_annot
self.img_path = '../data/InterHand2.6M/images'
self.annot_path = '../data/InterHand2.6M/annotations'
if self.annot_subset == 'machine_annot' and self.mode == 'val':
self.rootnet_output_path = '../data/InterHand2.6M/rootnet_output/rootnet_interhand2.6m_output_machine_annot_val.json'
else:
self.rootnet_output_path = '../data/InterHand2.6M/rootnet_output/rootnet_interhand2.6m_output_all_test.json'
self.transform = transform
self.joint_num = 21 # single hand
self.root_joint_idx = {'right': 20, 'left': 41}
self.joint_type = {'right': np.arange(0,self.joint_num), 'left': np.arange(self.joint_num,self.joint_num*2)}
self.skeleton = load_skeleton(osp.join(self.annot_path, 'skeleton.txt'), self.joint_num*2)
self.datalist = []
self.datalist_sh = []
self.datalist_ih = []
self.sequence_names = []
# load annotation
print("Load annotation from " + osp.join(self.annot_path, self.annot_subset))
db = COCO(osp.join(self.annot_path, self.annot_subset, 'InterHand2.6M_' + self.mode + '_data.json'))
with open(osp.join(self.annot_path, self.annot_subset, 'InterHand2.6M_' + self.mode + '_camera.json')) as f:
cameras = json.load(f)
with open(osp.join(self.annot_path, self.annot_subset, 'InterHand2.6M_' + self.mode + '_joint_3d.json')) as f:
joints = json.load(f)
if (self.mode == 'val' or self.mode == 'test') and cfg.trans_test == 'rootnet':
print("Get bbox and root depth from " + self.rootnet_output_path)
rootnet_result = {}
with open(self.rootnet_output_path) as f:
annot = json.load(f)
for i in range(len(annot)):
rootnet_result[str(annot[i]['annot_id'])] = annot[i]
else:
print("Get bbox and root depth from groundtruth annotation")
for aid in db.anns.keys():
ann = db.anns[aid]
image_id = ann['image_id']
img = db.loadImgs(image_id)[0]
capture_id = img['capture']
seq_name = img['seq_name']
cam = img['camera']
frame_idx = img['frame_idx']
img_path = osp.join(self.img_path, self.mode, img['file_name'])
campos, camrot = np.array(cameras[str(capture_id)]['campos'][str(cam)], dtype=np.float32), np.array(cameras[str(capture_id)]['camrot'][str(cam)], dtype=np.float32)
focal, princpt = np.array(cameras[str(capture_id)]['focal'][str(cam)], dtype=np.float32), np.array(cameras[str(capture_id)]['princpt'][str(cam)], dtype=np.float32)
joint_world = np.array(joints[str(capture_id)][str(frame_idx)], dtype=np.float32)
joint_cam = world2cam(joint_world.transpose(1,0), camrot, campos.reshape(3,1)).transpose(1,0)
joint_img = cam2pixel(joint_cam, focal, princpt)[:,:2]
joint_valid = np.array(ann['joint_valid'],dtype=np.float32).reshape(self.joint_num*2)
# if root is not valid -> root-relative 3D pose is also not valid. Therefore, mark all joints as invalid
joint_valid[self.joint_type['right']] *= joint_valid[self.root_joint_idx['right']]
joint_valid[self.joint_type['left']] *= joint_valid[self.root_joint_idx['left']]
hand_type = ann['hand_type']
hand_type_valid = np.array((ann['hand_type_valid']), dtype=np.float32)
if (self.mode == 'val' or self.mode == 'test') and cfg.trans_test == 'rootnet':
bbox = np.array(rootnet_result[str(aid)]['bbox'],dtype=np.float32)
abs_depth = {'right': rootnet_result[str(aid)]['abs_depth'][0], 'left': rootnet_result[str(aid)]['abs_depth'][1]}
else:
img_width, img_height = img['width'], img['height']
bbox = np.array(ann['bbox'],dtype=np.float32) # x,y,w,h
bbox = process_bbox(bbox, (img_height, img_width))
abs_depth = {'right': joint_cam[self.root_joint_idx['right'],2], 'left': joint_cam[self.root_joint_idx['left'],2]}
cam_param = {'focal': focal, 'princpt': princpt}
joint = {'cam_coord': joint_cam, 'img_coord': joint_img, 'valid': joint_valid}
data = {'img_path': img_path, 'seq_name': seq_name, 'cam_param': cam_param, 'bbox': bbox, 'joint': joint, 'hand_type': hand_type, 'hand_type_valid': hand_type_valid, 'abs_depth': abs_depth, 'file_name': img['file_name'], 'capture': capture_id, 'cam': cam, 'frame': frame_idx}
if hand_type == 'right' or hand_type == 'left':
self.datalist_sh.append(data)
else:
self.datalist_ih.append(data)
if seq_name not in self.sequence_names:
self.sequence_names.append(seq_name)
self.datalist = self.datalist_sh + self.datalist_ih
print('Number of annotations in single hand sequences: ' + str(len(self.datalist_sh)))
print('Number of annotations in interacting hand sequences: ' + str(len(self.datalist_ih)))
def __getitem__(self, idx):
data = copy.deepcopy(self.datalist[idx])
img_path, img_shape, bbox, joint_cam, cam_param, mano_param = data[
'img_path'], data['img_shape'], data['bbox'], data[
'joint_cam'], data['cam_param'], data['mano_param']
# img
img = load_img(img_path)
img, img2bb_trans, bb2img_trans, rot, _ = augmentation(
img, bbox, self.data_split, exclude_flip=True
) # FreiHAND dataset only contains right hands. do not perform flip aug.
img = self.transform(img.astype(np.float32)) / 255.
if self.data_split == 'train':
# mano coordinates
mano_mesh_cam, mano_joint_cam, mano_pose, mano_shape = self.get_mano_coord(
mano_param, cam_param)
mano_coord_cam = np.concatenate((mano_mesh_cam, mano_joint_cam))
focal, princpt = cam_param['focal'], cam_param['princpt']
mano_coord_img = cam2pixel(mano_coord_cam, focal, princpt)
# affine transform x,y coordinates. root-relative depth
mano_coord_img_xy1 = np.concatenate(
(mano_coord_img[:, :2], np.ones_like(mano_coord_img[:, :1])),
1)
mano_coord_img[:, :2] = np.dot(img2bb_trans,
mano_coord_img_xy1.transpose(
1, 0)).transpose(1, 0)[:, :2]
root_joint_depth = mano_coord_cam[self.vertex_num +
self.root_joint_idx][2]
mano_coord_img[:, 2] = mano_coord_img[:, 2] - root_joint_depth
mano_coord_img[:, 0] = mano_coord_img[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
mano_coord_img[:, 1] = mano_coord_img[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
mano_coord_img[:, 2] = (mano_coord_img[:, 2] /
(cfg.bbox_3d_size / 2) +
1) / 2. * cfg.output_hm_shape[0]
# check truncation
mano_trunc = ((mano_coord_img[:,0] >= 0) * (mano_coord_img[:,0] < cfg.output_hm_shape[2]) * \
(mano_coord_img[:,1] >= 0) * (mano_coord_img[:,1] < cfg.output_hm_shape[1]) * \
(mano_coord_img[:,2] >= 0) * (mano_coord_img[:,2] < cfg.output_hm_shape[0])).reshape(-1,1).astype(np.float32)
# split mesh and joint coordinates
mano_mesh_img = mano_coord_img[:self.vertex_num]
mano_joint_img = mano_coord_img[self.vertex_num:]
mano_mesh_trunc = mano_trunc[:self.vertex_num]
mano_joint_trunc = mano_trunc[self.vertex_num:]
# 3D data rotation augmentation
rot_aug_mat = np.array(
[[np.cos(np.deg2rad(-rot)), -np.sin(np.deg2rad(-rot)), 0],
[np.sin(np.deg2rad(-rot)),
np.cos(np.deg2rad(-rot)), 0], [0, 0, 1]],
dtype=np.float32)
# parameter
mano_pose = mano_pose.reshape(-1, 3)
root_pose = mano_pose[self.root_joint_idx, :]
root_pose, _ = cv2.Rodrigues(root_pose)
root_pose, _ = cv2.Rodrigues(np.dot(rot_aug_mat, root_pose))
mano_pose[self.root_joint_idx] = root_pose.reshape(3)
mano_pose = mano_pose.reshape(-1)
# mano coordinate
mano_joint_cam = mano_joint_cam - mano_joint_cam[
self.root_joint_idx, None] # root-relative
mano_joint_cam = np.dot(rot_aug_mat, mano_joint_cam.transpose(
1, 0)).transpose(1, 0)
orig_joint_img = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
orig_joint_cam = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
orig_joint_valid = np.zeros((self.joint_num, 1),
dtype=np.float32) # dummy
orig_joint_trunc = np.zeros((self.joint_num, 1),
dtype=np.float32) # dummy
inputs = {'img': img}
targets = {
'orig_joint_img': orig_joint_img,
'fit_joint_img': mano_joint_img,
'fit_mesh_img': mano_mesh_img,
'orig_joint_cam': orig_joint_cam,
'fit_joint_cam': mano_joint_cam,
'pose_param': mano_pose,
'shape_param': mano_shape
}
meta_info = {
'orig_joint_valid': orig_joint_valid,
'orig_joint_trunc': orig_joint_trunc,
'fit_joint_trunc': mano_joint_trunc,
'fit_mesh_trunc': mano_mesh_trunc,
'is_valid_fit': float(True),
'is_3D': float(True)
}
else:
inputs = {'img': img}
targets = {}
meta_info = {'bb2img_trans': bb2img_trans}
return inputs, targets, meta_info
root_depth = 11250.5732421875 # obtain this from RootNet (https://github.com/mks0601/3DMPPE_ROOTNET_RELEASE/tree/master/demo)
root_depth /= 1000 # output of RootNet is milimeter. change it to meter
root_img = np.array([root_xy[0], root_xy[1], root_depth])
root_cam = pixel2cam(root_img[None,:], focal, princpt)
mesh_lixel_img[:,2] += root_depth
mesh_lixel_cam = pixel2cam(mesh_lixel_img, focal, princpt)
mesh_param_cam += root_cam.reshape(1,3)
# visualize lixel mesh in 2D space
vis_img = original_img.copy()
vis_img = vis_mesh(vis_img, mesh_lixel_img)
cv2.imwrite('output_mesh_lixel.jpg', vis_img)
# visualize lixel mesh in 2D space
vis_img = original_img.copy()
mesh_param_img = cam2pixel(mesh_param_cam, focal, princpt)
vis_img = vis_mesh(vis_img, mesh_param_img)
cv2.imwrite('output_mesh_param.jpg', vis_img)
# save mesh (obj)
save_obj(mesh_lixel_cam, face, 'output_mesh_lixel.obj')
save_obj(mesh_param_cam, face, 'output_mesh_param.obj')
# render mesh from lixel
vis_img = original_img.copy()
rendered_img = render_mesh(vis_img, mesh_lixel_cam, face, {'focal': focal, 'princpt': princpt})
cv2.imwrite('rendered_mesh_lixel.jpg', rendered_img)
# render mesh from param
vis_img = original_img.copy()
rendered_img = render_mesh(vis_img, mesh_param_cam, face, {'focal': focal, 'princpt': princpt})
def __getitem__(self, idx):
data = copy.deepcopy(self.datalist[idx])
img_path, img_shape, bbox, smpl_param, cam_param = data[
'img_path'], data['img_shape'], data['bbox'], data[
'smpl_param'], data['cam_param']
# img
img = load_img(img_path)
img, img2bb_trans, bb2img_trans, rot, do_flip = augmentation(
img, bbox, self.data_split)
img = self.transform(img.astype(np.float32)) / 255.
# muco gt
muco_joint_img = data['joint_img']
muco_joint_cam = data['joint_cam']
muco_joint_cam = muco_joint_cam - muco_joint_cam[
self.muco_root_joint_idx, None, :] # root-relative
muco_joint_valid = data['joint_valid']
if do_flip:
muco_joint_img[:, 0] = img_shape[1] - 1 - muco_joint_img[:, 0]
muco_joint_cam[:, 0] = -muco_joint_cam[:, 0]
for pair in self.muco_flip_pairs:
muco_joint_img[pair[0], :], muco_joint_img[
pair[1], :] = muco_joint_img[
pair[1], :].copy(), muco_joint_img[pair[0], :].copy()
muco_joint_cam[pair[0], :], muco_joint_cam[
pair[1], :] = muco_joint_cam[
pair[1], :].copy(), muco_joint_cam[pair[0], :].copy()
muco_joint_valid[pair[0], :], muco_joint_valid[
pair[1], :] = muco_joint_valid[pair[1], :].copy(
), muco_joint_valid[pair[0], :].copy()
muco_joint_img_xy1 = np.concatenate(
(muco_joint_img[:, :2], np.ones_like(muco_joint_img[:, :1])), 1)
muco_joint_img[:, :2] = np.dot(img2bb_trans,
muco_joint_img_xy1.transpose(
1, 0)).transpose(1, 0)
muco_joint_img[:, 0] = muco_joint_img[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
muco_joint_img[:, 1] = muco_joint_img[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
muco_joint_img[:, 2] = muco_joint_img[:, 2] - muco_joint_img[
self.muco_root_joint_idx][2] # root-relative
muco_joint_img[:, 2] = (
muco_joint_img[:, 2] /
(cfg.bbox_3d_size * 1000 / 2) + 1) / 2. * cfg.output_hm_shape[
0] # change cfg.bbox_3d_size from meter to milimeter
# check truncation
muco_joint_trunc = muco_joint_valid * ((muco_joint_img[:,0] >= 0) * (muco_joint_img[:,0] < cfg.output_hm_shape[2]) * \
(muco_joint_img[:,1] >= 0) * (muco_joint_img[:,1] < cfg.output_hm_shape[1]) * \
(muco_joint_img[:,2] >= 0) * (muco_joint_img[:,2] < cfg.output_hm_shape[0])).reshape(-1,1).astype(np.float32)
# transform muco joints to target db joints
muco_joint_img = transform_joint_to_other_db(muco_joint_img,
self.muco_joints_name,
self.joints_name)
muco_joint_cam = transform_joint_to_other_db(muco_joint_cam,
self.muco_joints_name,
self.joints_name)
muco_joint_valid = transform_joint_to_other_db(muco_joint_valid,
self.muco_joints_name,
self.joints_name)
muco_joint_trunc = transform_joint_to_other_db(muco_joint_trunc,
self.muco_joints_name,
self.joints_name)
if smpl_param is not None:
# smpl coordinates
smpl_mesh_cam, smpl_joint_cam, smpl_pose, smpl_shape = self.get_smpl_coord(
smpl_param, cam_param, do_flip, img_shape)
smpl_coord_cam = np.concatenate((smpl_mesh_cam, smpl_joint_cam))
focal, princpt = cam_param['focal'], cam_param['princpt']
smpl_coord_img = cam2pixel(smpl_coord_cam, focal, princpt)
# affine transform x,y coordinates. root-relative depth
smpl_coord_img_xy1 = np.concatenate(
(smpl_coord_img[:, :2], np.ones_like(smpl_coord_img[:, :1])),
1)
smpl_coord_img[:, :2] = np.dot(img2bb_trans,
smpl_coord_img_xy1.transpose(
1, 0)).transpose(1, 0)[:, :2]
smpl_coord_img[:, 2] = smpl_coord_img[:, 2] - smpl_coord_cam[
self.vertex_num + self.root_joint_idx][2]
smpl_coord_img[:, 0] = smpl_coord_img[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
smpl_coord_img[:, 1] = smpl_coord_img[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
smpl_coord_img[:, 2] = (
smpl_coord_img[:, 2] /
(cfg.bbox_3d_size * 1000 / 2) + 1) / 2. * cfg.output_hm_shape[
0] # change cfg.bbox_3d_size from meter to milimeter
# check truncation
smpl_trunc = ((smpl_coord_img[:,0] >= 0) * (smpl_coord_img[:,0] < cfg.output_hm_shape[2]) * \
(smpl_coord_img[:,1] >= 0) * (smpl_coord_img[:,1] < cfg.output_hm_shape[1]) * \
(smpl_coord_img[:,2] >= 0) * (smpl_coord_img[:,2] < cfg.output_hm_shape[0])).reshape(-1,1).astype(np.float32)
# split mesh and joint coordinates
smpl_mesh_img = smpl_coord_img[:self.vertex_num]
smpl_joint_img = smpl_coord_img[self.vertex_num:]
smpl_mesh_trunc = smpl_trunc[:self.vertex_num]
smpl_joint_trunc = smpl_trunc[self.vertex_num:]
# if fitted mesh is too far from muco gt, discard it
is_valid_fit = True
error = self.get_fitting_error(data['joint_cam'], smpl_mesh_cam,
do_flip)
if error > self.fitting_thr:
is_valid_fit = False
else:
smpl_joint_img = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
smpl_joint_cam = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
smpl_mesh_img = np.zeros((self.vertex_num, 3),
dtype=np.float32) # dummy
smpl_pose = np.zeros((72), dtype=np.float32) # dummy
smpl_shape = np.zeros((10), dtype=np.float32) # dummy
smpl_joint_trunc = np.zeros((self.joint_num, 1),
dtype=np.float32) # dummy
smpl_mesh_trunc = np.zeros((self.vertex_num, 1),
dtype=np.float32) # dummy
is_valid_fit = False
# 3D data rotation augmentation
rot_aug_mat = np.array(
[[np.cos(np.deg2rad(-rot)), -np.sin(np.deg2rad(-rot)), 0],
[np.sin(np.deg2rad(-rot)),
np.cos(np.deg2rad(-rot)), 0], [0, 0, 1]],
dtype=np.float32)
# muco coordinate
muco_joint_cam = np.dot(rot_aug_mat, muco_joint_cam.transpose(
1, 0)).transpose(1, 0) / 1000 # milimeter to meter
# parameter
smpl_pose = smpl_pose.reshape(-1, 3)
root_pose = smpl_pose[self.root_joint_idx, :]
root_pose, _ = cv2.Rodrigues(root_pose)
root_pose, _ = cv2.Rodrigues(np.dot(rot_aug_mat, root_pose))
smpl_pose[self.root_joint_idx] = root_pose.reshape(3)
smpl_pose = smpl_pose.reshape(-1)
# smpl coordinate
smpl_joint_cam = smpl_joint_cam - smpl_joint_cam[self.root_joint_idx,
None] # root-relative
smpl_joint_cam = np.dot(rot_aug_mat, smpl_joint_cam.transpose(
1, 0)).transpose(1, 0) / 1000 # milimeter to meter
inputs = {'img': img}
targets = {
'orig_joint_img': muco_joint_img,
'fit_joint_img': smpl_joint_img,
'fit_mesh_img': smpl_mesh_img,
'orig_joint_cam': muco_joint_cam,
'fit_joint_cam': smpl_joint_cam,
'pose_param': smpl_pose,
'shape_param': smpl_shape
}
meta_info = {
'orig_joint_valid': muco_joint_valid,
'orig_joint_trunc': muco_joint_trunc,
'fit_joint_trunc': smpl_joint_trunc,
'fit_mesh_trunc': smpl_mesh_trunc,
'is_valid_fit': float(is_valid_fit),
'is_3D': float(True)
}
return inputs, targets, meta_info
def __getitem__(self, idx):
data = copy.deepcopy(self.datalist[idx])
img_path, img_shape, bbox = data['img_path'], data['img_shape'], data[
'bbox']
# image load and affine transform
img = load_img(img_path)
img, img2bb_trans, bb2img_trans, rot, do_flip = augmentation(
img, bbox, self.data_split)
img = self.transform(img.astype(np.float32)) / 255.
if self.data_split == 'train':
# coco gt
coco_joint_img = data['joint_img']
coco_joint_valid = data['joint_valid']
if do_flip:
coco_joint_img[:, 0] = img_shape[1] - 1 - coco_joint_img[:, 0]
for pair in self.coco_flip_pairs:
coco_joint_img[pair[0], :], coco_joint_img[
pair[1], :] = coco_joint_img[pair[1], :].copy(
), coco_joint_img[pair[0], :].copy()
coco_joint_valid[pair[0], :], coco_joint_valid[
pair[1], :] = coco_joint_valid[pair[1], :].copy(
), coco_joint_valid[pair[0], :].copy()
coco_joint_img_xy1 = np.concatenate(
(coco_joint_img[:, :2], np.ones_like(coco_joint_img[:, :1])),
1)
coco_joint_img[:, :2] = np.dot(img2bb_trans,
coco_joint_img_xy1.transpose(
1, 0)).transpose(1, 0)
coco_joint_img[:, 0] = coco_joint_img[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
coco_joint_img[:, 1] = coco_joint_img[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
# backup for calculating fitting error
_coco_joint_img = coco_joint_img.copy()
_coco_joint_valid = coco_joint_valid.copy()
# check truncation
coco_joint_trunc = coco_joint_valid * ((coco_joint_img[:,0] >= 0) * (coco_joint_img[:,0] < cfg.output_hm_shape[2]) * \
(coco_joint_img[:,1] >= 0) * (coco_joint_img[:,1] < cfg.output_hm_shape[1])).reshape(-1,1).astype(np.float32)
# transform coco joints to target db joints
coco_joint_img = transform_joint_to_other_db(
coco_joint_img, self.coco_joints_name, self.joints_name)
coco_joint_cam = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
coco_joint_valid = transform_joint_to_other_db(
coco_joint_valid, self.coco_joints_name, self.joints_name)
coco_joint_trunc = transform_joint_to_other_db(
coco_joint_trunc, self.coco_joints_name, self.joints_name)
smplify_result = data['smplify_result']
if smplify_result is not None:
# use fitted mesh
smpl_param, cam_param = smplify_result[
'smpl_param'], smplify_result['cam_param']
smpl_mesh_cam, smpl_joint_cam, smpl_pose, smpl_shape = self.get_smpl_coord(
smpl_param, cam_param, do_flip, img_shape)
smpl_coord_cam = np.concatenate(
(smpl_mesh_cam, smpl_joint_cam))
smpl_coord_img = cam2pixel(smpl_coord_cam, cam_param['focal'],
cam_param['princpt'])
# x,y affine transform, root-relative depth
smpl_coord_img_xy1 = np.concatenate(
(smpl_coord_img[:, :2], np.ones_like(
smpl_coord_img[:, 0:1])), 1)
smpl_coord_img[:, :2] = np.dot(
img2bb_trans,
smpl_coord_img_xy1.transpose(1, 0)).transpose(1, 0)[:, :2]
smpl_coord_img[:, 2] = smpl_coord_img[:, 2] - smpl_coord_cam[
self.vertex_num + self.root_joint_idx][2]
smpl_coord_img[:,
0] = smpl_coord_img[:, 0] / cfg.input_img_shape[
1] * cfg.output_hm_shape[2]
smpl_coord_img[:,
1] = smpl_coord_img[:, 1] / cfg.input_img_shape[
0] * cfg.output_hm_shape[1]
smpl_coord_img[:, 2] = (smpl_coord_img[:, 2] /
(cfg.bbox_3d_size / 2) +
1) / 2. * cfg.output_hm_shape[0]
# check truncation
smpl_trunc = ((smpl_coord_img[:,0] >= 0) * (smpl_coord_img[:,0] < cfg.output_hm_shape[2]) * \
(smpl_coord_img[:,1] >= 0) * (smpl_coord_img[:,1] < cfg.output_hm_shape[1]) * \
(smpl_coord_img[:,2] >= 0) * (smpl_coord_img[:,2] < cfg.output_hm_shape[0])).reshape(-1,1).astype(np.float32)
# split mesh and joint coordinates
smpl_mesh_img = smpl_coord_img[:self.vertex_num]
smpl_joint_img = smpl_coord_img[self.vertex_num:]
smpl_mesh_trunc = smpl_trunc[:self.vertex_num]
smpl_joint_trunc = smpl_trunc[self.vertex_num:]
# if fitted mesh is too far from h36m gt, discard it
is_valid_fit = True
error = self.get_fitting_error(_coco_joint_img, smpl_mesh_cam,
cam_param, img2bb_trans,
_coco_joint_valid)
if error > self.fitting_thr:
is_valid_fit = False
else:
smpl_joint_img = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
smpl_joint_cam = np.zeros((self.joint_num, 3),
dtype=np.float32) # dummy
smpl_mesh_img = np.zeros((self.vertex_num, 3),
dtype=np.float32) # dummy
smpl_pose = np.zeros((72), dtype=np.float32) # dummy
smpl_shape = np.zeros((10), dtype=np.float32) # dummy
smpl_joint_trunc = np.zeros((self.joint_num, 1),
dtype=np.float32)
smpl_mesh_trunc = np.zeros((self.vertex_num, 1),
dtype=np.float32)
is_valid_fit = False
# 3D data rotation augmentation
rot_aug_mat = np.array(
[[np.cos(np.deg2rad(-rot)), -np.sin(np.deg2rad(-rot)), 0],
[np.sin(np.deg2rad(-rot)),
np.cos(np.deg2rad(-rot)), 0], [0, 0, 1]],
dtype=np.float32)
# parameter
smpl_pose = smpl_pose.reshape(-1, 3)
root_pose = smpl_pose[self.root_joint_idx, :]
root_pose, _ = cv2.Rodrigues(root_pose)
root_pose, _ = cv2.Rodrigues(np.dot(rot_aug_mat, root_pose))
smpl_pose[self.root_joint_idx] = root_pose.reshape(3)
smpl_pose = smpl_pose.reshape(-1)
# smpl coordinate
smpl_joint_cam = smpl_joint_cam - smpl_joint_cam[
self.root_joint_idx, None] # root-relative
smpl_joint_cam = np.dot(rot_aug_mat, smpl_joint_cam.transpose(
1, 0)).transpose(1, 0)
inputs = {'img': img}
targets = {
'orig_joint_img': coco_joint_img,
'fit_joint_img': smpl_joint_img,
'fit_mesh_img': smpl_mesh_img,
'orig_joint_cam': coco_joint_cam,
'fit_joint_cam': smpl_joint_cam,
'pose_param': smpl_pose,
'shape_param': smpl_shape
}
meta_info = {
'orig_joint_valid': coco_joint_valid,
'orig_joint_trunc': coco_joint_trunc,
'fit_joint_trunc': smpl_joint_trunc,
'fit_mesh_trunc': smpl_mesh_trunc,
'is_valid_fit': float(is_valid_fit),
'is_3D': float(False)
}
return inputs, targets, meta_info
else:
inputs = {'img': img}
targets = {}
meta_info = {'bb2img_trans': bb2img_trans}
return inputs, targets, meta_info
def load_data(self):
subject_list = self.get_subject()
sampling_ratio = self.get_subsampling_ratio()
# aggregate annotations from each subject
db = COCO()
cameras = {}
joints = {}
smpl_params = {}
for subject in subject_list:
# data load
with open(
osp.join(self.annot_path,
'Human36M_subject' + str(subject) + '_data.json'),
'r') as f:
annot = json.load(f)
if len(db.dataset) == 0:
for k, v in annot.items():
db.dataset[k] = v
else:
for k, v in annot.items():
db.dataset[k] += v
# camera load
with open(
osp.join(
self.annot_path,
'Human36M_subject' + str(subject) + '_camera.json'),
'r') as f:
cameras[str(subject)] = json.load(f)
# joint coordinate load
with open(
osp.join(
self.annot_path,
'Human36M_subject' + str(subject) + '_joint_3d.json'),
'r') as f:
joints[str(subject)] = json.load(f)
# smpl parameter load
with open(
osp.join(
self.annot_path, 'Human36M_subject' + str(subject) +
'_smpl_param.json'), 'r') as f:
smpl_params[str(subject)] = json.load(f)
db.createIndex()
if self.data_split == 'test' and not cfg.use_gt_info:
print("Get bounding box and root from " + self.human_bbox_root_dir)
bbox_root_result = {}
with open(self.human_bbox_root_dir) as f:
annot = json.load(f)
for i in range(len(annot)):
bbox_root_result[str(annot[i]['image_id'])] = {
'bbox': np.array(annot[i]['bbox']),
'root': np.array(annot[i]['root_cam'])
}
else:
print("Get bounding box and root from groundtruth")
datalist = []
for aid in db.anns.keys():
ann = db.anns[aid]
image_id = ann['image_id']
img = db.loadImgs(image_id)[0]
img_path = osp.join(self.img_dir, img['file_name'])
img_shape = (img['height'], img['width'])
# check subject and frame_idx
frame_idx = img['frame_idx']
if frame_idx % sampling_ratio != 0:
continue
# check smpl parameter exist
subject = img['subject']
action_idx = img['action_idx']
subaction_idx = img['subaction_idx']
frame_idx = img['frame_idx']
try:
smpl_param = smpl_params[str(subject)][str(action_idx)][str(
subaction_idx)][str(frame_idx)]
except KeyError:
smpl_param = None
# camera parameter
cam_idx = img['cam_idx']
cam_param = cameras[str(subject)][str(cam_idx)]
R, t, f, c = np.array(cam_param['R'], dtype=np.float32), np.array(
cam_param['t'], dtype=np.float32), np.array(
cam_param['f'],
dtype=np.float32), np.array(cam_param['c'],
dtype=np.float32)
cam_param = {'R': R, 't': t, 'focal': f, 'princpt': c}
# only use frontal camera following previous works (HMR and SPIN)
if self.data_split == 'test' and str(cam_idx) != '4':
continue
# project world coordinate to cam, image coordinate space
joint_world = np.array(joints[str(subject)][str(action_idx)][str(
subaction_idx)][str(frame_idx)],
dtype=np.float32)
joint_cam = world2cam(joint_world, R, t)
joint_img = cam2pixel(joint_cam, f, c)
joint_valid = np.ones((self.h36m_joint_num, 1))
if self.data_split == 'test' and not cfg.use_gt_info:
bbox = bbox_root_result[str(
image_id
)]['bbox'] # bbox should be aspect ratio preserved-extended. It is done in RootNet.
root_joint_depth = bbox_root_result[str(image_id)]['root'][2]
else:
bbox = process_bbox(np.array(ann['bbox']), img['width'],
img['height'])
if bbox is None: continue
root_joint_depth = joint_cam[self.h36m_root_joint_idx][2]
datalist.append({
'img_path': img_path,
'img_id': image_id,
'img_shape': img_shape,
'bbox': bbox,
'joint_img': joint_img,
'joint_cam': joint_cam,
'joint_valid': joint_valid,
'smpl_param': smpl_param,
'root_joint_depth': root_joint_depth,
'cam_param': cam_param
})
return datalist
