def transfer_hand_wrist(
smplx_model, body_pose, hand_wrist, hand_type,
transfer_type="g2l", result_format="rotmat"):
assert transfer_type in ["g2l", "l2g"]
assert result_format in ['rotmat', 'aa']
if hand_type == 'left_hand':
kinematic_map = get_kinematic_map(smplx_model, 20)
else:
assert hand_type == 'right_hand'
kinematic_map = get_kinematic_map(smplx_model, 21)
if transfer_type == "l2g":
# local to global
hand_wrist_local = hand_wrist.clone()
hand_wrist_mat = __transfer_hand_rot(
body_pose, hand_wrist_local, kinematic_map, transfer_type)
else:
# global to local
assert transfer_type == "g2l"
hand_wrist_global = hand_wrist.clone()
hand_wrist_mat = __transfer_hand_rot(
body_pose, hand_wrist_global, kinematic_map, transfer_type)
if result_format == 'rotmat':
return hand_wrist_mat
else:
hand_wrist_aa = gu.rotation_matrix_to_angle_axis(hand_wrist_mat)
return hand_wrist_aa
def transfer_rotation(smplx_model,
body_pose,
part_rot,
part_idx,
transfer_type="g2l",
result_format="rotmat"):
assert transfer_type in ["g2l", "l2g"]
assert result_format in ['rotmat', 'aa']
assert type(body_pose) == type(part_rot)
return_np = False
if isinstance(body_pose, np.ndarray):
body_pose = torch.from_numpy(body_pose)
return_np = True
if isinstance(part_rot, np.ndarray):
part_rot = torch.from_numpy(part_rot)
return_np = True
if body_pose.dim() == 2:
# aa
assert body_pose.size(0) == 1 and body_pose.size(1) in [66, 72]
body_pose_rotmat = gu.angle_axis_to_rotation_matrix(
body_pose.view(22, 3)).clone()
else:
# rotmat
assert body_pose.dim() == 4
assert body_pose.size(0) == 1 and body_pose.size(1) in [22, 24]
assert body_pose.size(2) == 3 and body_pose.size(3) == 3
body_pose_rotmat = body_pose[0].clone()
if part_rot.dim() == 2:
# aa
assert part_rot.size(0) == 1 and part_rot.size(1) == 3
part_rotmat = gu.angle_axis_to_rotation_matrix(part_rot)[
0, :3, :3].clone()
else:
# rotmat
assert part_rot.dim() == 3
assert part_rot.size(0) == 1 and part_rot.size(
1) == 3 and part_rot.size(2) == 3
part_rotmat = part_rot[0, :3, :3].clone()
kinematic_map = get_kinematic_map(smplx_model, part_idx)
part_rot_trans = __transfer_rot(body_pose_rotmat, part_rotmat,
kinematic_map, transfer_type)
if result_format == 'rotmat':
return_value = part_rot_trans
else:
part_rot_aa = gu.rotation_matrix_to_angle_axis(part_rot_trans)
return_value = part_rot_aa
if return_np:
return_value = return_value.numpy()
return return_value
def get_local_hand_rot(body_pose, hand_rot_global, kinematic_map):
hand_rotmat_global = gu.angle_axis_to_rotation_matrix(
hand_rot_global.view(1, 3))
body_pose = body_pose.reshape(-1, 3)
# the shape is (1,4,4), torch matmul support 3 dimension
rotmat = gu.angle_axis_to_rotation_matrix(body_pose[0].view(1, 3))
parent_id = 0
while parent_id in kinematic_map:
child_id = kinematic_map[parent_id]
local_rotmat = gu.angle_axis_to_rotation_matrix(
body_pose[child_id].view(1, 3))
rotmat = torch.matmul(rotmat, local_rotmat)
parent_id = child_id
hand_rotmat_local = torch.matmul(rotmat.inverse(), hand_rotmat_global)
# print("hand_rotmat_local", hand_rotmat_local.size())
hand_rot_local = gu.rotation_matrix_to_angle_axis(
hand_rotmat_local[:, :3, :])
return hand_rot_local
def integration_copy_paste(pred_body_list, pred_hand_list, smplx_model, image_shape):
integral_output_list = list()
for i in range(len(pred_body_list)):
body_info = pred_body_list[i]
hand_info = pred_hand_list[i]
if body_info is None:
integral_output_list.append(None)
continue
# copy and paste
pred_betas = torch.from_numpy(body_info['pred_betas']).cuda()
pred_rotmat = torch.from_numpy(body_info['pred_rotmat']).cuda()
# integrate right hand pose
hand_output = dict()
if hand_info is not None and hand_info['right_hand'] is not None:
right_hand_pose = torch.from_numpy(hand_info['right_hand']['pred_hand_pose'][:, 3:]).cuda()
right_hand_global_orient = torch.from_numpy(hand_info['right_hand']['pred_hand_pose'][:,: 3]).cuda()
right_hand_local_orient = transfer_rotation(
smplx_model, pred_rotmat, right_hand_global_orient, 21)
pred_rotmat[0, 21] = right_hand_local_orient
else:
right_hand_pose = torch.from_numpy(np.zeros( (1,45) , dtype= np.float32)).cuda()
right_hand_global_orient = None
right_hand_local_orient = None
# integrate left hand pose
if hand_info is not None and hand_info['left_hand'] is not None:
left_hand_pose = torch.from_numpy(hand_info['left_hand']['pred_hand_pose'][:, 3:]).cuda()
left_hand_global_orient = torch.from_numpy(hand_info['left_hand']['pred_hand_pose'][:, :3]).cuda()
left_hand_local_orient = transfer_rotation(
smplx_model, pred_rotmat, left_hand_global_orient, 20)
pred_rotmat[0, 20] = left_hand_local_orient
else:
left_hand_pose = torch.from_numpy(np.zeros((1,45), dtype= np.float32)).cuda()
left_hand_global_orient = None
left_hand_local_orient = None
pred_aa = gu.rotation_matrix_to_angle_axis(pred_rotmat).cuda()
pred_aa = pred_aa.reshape(pred_aa.shape[0], 72)
smplx_output = smplx_model(
betas = pred_betas,
body_pose = pred_aa[:,3:],
global_orient = pred_aa[:,:3],
right_hand_pose = right_hand_pose,
left_hand_pose= left_hand_pose,
pose2rot = True)
pred_vertices = smplx_output.vertices
pred_vertices = pred_vertices[0].detach().cpu().numpy()
pred_joints_3d = smplx_output.joints
pred_joints_3d = pred_joints_3d[0].detach().cpu().numpy()
camScale = body_info['pred_camera'][0]
camTrans = body_info['pred_camera'][1:]
bbox_top_left = body_info['bbox_top_left']
bbox_scale_ratio = body_info['bbox_scale_ratio']
integral_output = dict()
integral_output['pred_vertices_smpl'] = pred_vertices
integral_output['faces'] = smplx_model.faces
integral_output['bbox_scale_ratio'] = bbox_scale_ratio
integral_output['bbox_top_left'] = bbox_top_left
integral_output['pred_camera'] = body_info['pred_camera']
pred_aa_tensor = gu.rotation_matrix_to_angle_axis(pred_rotmat.detach().cpu()[0])
integral_output['pred_body_pose'] = pred_aa_tensor.cpu().numpy().reshape(1, 72)
integral_output['pred_betas'] = pred_betas.detach().cpu().numpy()
# convert mesh to original image space (X,Y are aligned to image)
pred_vertices_bbox = convert_smpl_to_bbox(
pred_vertices, camScale, camTrans)
pred_vertices_img = convert_bbox_to_oriIm(
pred_vertices_bbox, bbox_scale_ratio, bbox_top_left, image_shape[1], image_shape[0])
integral_output['pred_vertices_img'] = pred_vertices_img
# convert mesh to original image space (X,Y are aligned to image)
pred_joints_bbox = convert_smpl_to_bbox(
pred_joints_3d, camScale, camTrans)
pred_joints_img = convert_bbox_to_oriIm(
pred_joints_bbox, bbox_scale_ratio, bbox_top_left, image_shape[1], image_shape[0])
integral_output['pred_joints_img'] = pred_joints_img
pred_left_hand_joints = smplx_output.left_hand_joints
pred_left_hand_joints = pred_left_hand_joints[0].detach().cpu().numpy()
# convert mesh to original image space (X,Y are aligned to image)
pred_left_hand_joints_bbox = convert_smpl_to_bbox(
pred_left_hand_joints, camScale, camTrans)
pred_left_hand_joints_img = convert_bbox_to_oriIm(
pred_left_hand_joints_bbox, bbox_scale_ratio, bbox_top_left, image_shape[1], image_shape[0])
integral_output['pred_left_hand_joints_img'] = pred_left_hand_joints_img
pred_right_hand_joints = smplx_output.right_hand_joints
pred_right_hand_joints = pred_right_hand_joints[0].detach().cpu().numpy()
# convert mesh to original image space (X,Y are aligned to image)
pred_right_hand_joints_bbox = convert_smpl_to_bbox(
pred_right_hand_joints, camScale, camTrans)
pred_right_hand_joints_img = convert_bbox_to_oriIm(
pred_right_hand_joints_bbox, bbox_scale_ratio, bbox_top_left, image_shape[1], image_shape[0])
integral_output['pred_right_hand_joints_img'] = pred_right_hand_joints_img
# keep hand info
r_hand_local_orient_body = body_info['pred_rotmat'][:, 21] # rot-mat
r_hand_global_orient_body = transfer_rotation(
smplx_model, pred_rotmat,
torch.from_numpy(r_hand_local_orient_body).cuda(),
21, 'l2g', 'aa').numpy().reshape(1, 3) # aa
r_hand_local_orient_body = gu.rotation_matrix_to_angle_axis(r_hand_local_orient_body) # rot-mat -> aa
l_hand_local_orient_body = body_info['pred_rotmat'][:, 20]
l_hand_global_orient_body = transfer_rotation(
smplx_model, pred_rotmat,
torch.from_numpy(l_hand_local_orient_body).cuda(),
20, 'l2g', 'aa').numpy().reshape(1, 3)
l_hand_local_orient_body = gu.rotation_matrix_to_angle_axis(l_hand_local_orient_body) # rot-mat -> aa
r_hand_local_orient_hand = None
r_hand_global_orient_hand = None
if right_hand_local_orient is not None:
r_hand_local_orient_hand = gu.rotation_matrix_to_angle_axis(
right_hand_local_orient).detach().cpu().numpy().reshape(1, 3)
r_hand_global_orient_hand = right_hand_global_orient.detach().cpu().numpy().reshape(1, 3)
l_hand_local_orient_hand = None
l_hand_global_orient_hand = None
if left_hand_local_orient is not None:
l_hand_local_orient_hand = gu.rotation_matrix_to_angle_axis(
left_hand_local_orient).detach().cpu().numpy().reshape(1, 3)
l_hand_global_orient_hand = left_hand_global_orient.detach().cpu().numpy().reshape(1, 3)
# poses and rotations related to hands
integral_output['left_hand_local_orient_body'] = l_hand_local_orient_body
integral_output['left_hand_global_orient_body'] = l_hand_global_orient_body
integral_output['right_hand_local_orient_body'] = r_hand_local_orient_body
integral_output['right_hand_global_orient_body'] = r_hand_global_orient_body
integral_output['left_hand_local_orient_hand'] = l_hand_local_orient_hand
integral_output['left_hand_global_orient_hand'] = l_hand_global_orient_hand
integral_output['right_hand_local_orient_hand'] = r_hand_local_orient_hand
integral_output['right_hand_global_orient_hand'] = r_hand_global_orient_hand
integral_output['pred_left_hand_pose'] = left_hand_pose.detach().cpu().numpy()
integral_output['pred_right_hand_pose'] = right_hand_pose.detach().cpu().numpy()
# predicted hand betas, cameras, top-left corner and center
left_hand_betas = None
left_hand_camera = None
left_hand_bbox_scale = None
left_hand_bbox_top_left = None
if hand_info is not None and hand_info['left_hand'] is not None:
left_hand_betas = hand_info['left_hand']['pred_hand_betas']
left_hand_camera = hand_info['left_hand']['pred_camera']
left_hand_bbox_scale = hand_info['left_hand']['bbox_scale_ratio']
left_hand_bbox_top_left = hand_info['left_hand']['bbox_top_left']
right_hand_betas = None
right_hand_camera = None
right_hand_bbox_scale = None
right_hand_bbox_top_left = None
if hand_info is not None and hand_info['right_hand'] is not None:
right_hand_betas = hand_info['right_hand']['pred_hand_betas']
right_hand_camera = hand_info['right_hand']['pred_camera']
right_hand_bbox_scale = hand_info['right_hand']['bbox_scale_ratio']
right_hand_bbox_top_left = hand_info['right_hand']['bbox_top_left']
integral_output['pred_left_hand_betas'] = left_hand_betas
integral_output['left_hand_camera'] = left_hand_camera
integral_output['left_hand_bbox_scale_ratio'] = left_hand_bbox_scale
integral_output['left_hand_bbox_top_left'] = left_hand_bbox_top_left
integral_output['pred_right_hand_betas'] = right_hand_betas
integral_output['right_hand_camera'] = right_hand_camera
integral_output['right_hand_bbox_scale_ratio'] = right_hand_bbox_scale
integral_output['right_hand_bbox_top_left'] = right_hand_bbox_top_left
integral_output_list.append(integral_output)
return integral_output_list
def regress(self, img_original, body_bbox_list):
"""
args:
img_original: original raw image (BGR order by using cv2.imread)
body_bbox: bounding box around the target: (minX, minY, width, height)
outputs:
pred_vertices_img:
pred_joints_vis_img:
pred_rotmat
pred_betas
pred_camera
bbox: [bbr[0], bbr[1],bbr[0]+bbr[2], bbr[1]+bbr[3]])
bboxTopLeft: bbox top left (redundant)
boxScale_o2n: bbox scaling factor (redundant)
"""
pred_output_list = list()
for body_bbox in body_bbox_list:
img, norm_img, boxScale_o2n, bboxTopLeft, bbox = process_image_bbox(
img_original, body_bbox, input_res=constants.IMG_RES)
# bboxTopLeft = bbox['bboxXYWH'][:2]
'''
print("bboxTopLeft", bboxTopLeft)
print('img', img.shape)
cv2.imwrite("img.png", img)
print("here !!!!", )
sys.exit(0)
'''
if img is None:
pred_output_list.append(None)
continue
with torch.no_grad():
# model forward
pred_rotmat, pred_betas, pred_camera = self.model_regressor(
norm_img.to(self.device))
#Convert rot_mat to aa since hands are always in aa
pred_aa = rotmat3x3_to_angleaxis(pred_rotmat)
pred_aa = pred_aa.view(pred_aa.shape[0], -1)
smpl_output = self.smpl(betas=pred_betas,
body_pose=pred_aa[:, 3:],
global_orient=pred_aa[:, :3],
pose2rot=True)
pred_vertices = smpl_output.vertices
pred_joints_3d = smpl_output.joints
pred_vertices = pred_vertices[0].cpu().numpy()
pred_camera = pred_camera.cpu().numpy().ravel()
camScale = pred_camera[0] # *1.15
camTrans = pred_camera[1:]
pred_output = dict()
# Convert mesh to original image space (X,Y are aligned to image)
# 1. SMPL -> 2D bbox
# 2. 2D bbox -> original 2D image
pred_vertices_bbox = convert_smpl_to_bbox(
pred_vertices, camScale, camTrans)
pred_vertices_img = convert_bbox_to_oriIm(
pred_vertices_bbox, boxScale_o2n, bboxTopLeft,
img_original.shape[1], img_original.shape[0])
# Convert joint to original image space (X,Y are aligned to image)
pred_joints_3d = pred_joints_3d[0].cpu().numpy() # (1,49,3)
pred_joints_vis = pred_joints_3d[:, :3] # (49,3)
pred_joints_vis_bbox = convert_smpl_to_bbox(
pred_joints_vis, camScale, camTrans)
pred_joints_vis_img = convert_bbox_to_oriIm(
pred_joints_vis_bbox, boxScale_o2n, bboxTopLeft,
img_original.shape[1], img_original.shape[0])
# Output
pred_output['img_cropped'] = img[:, :, ::-1]
pred_output['pred_vertices_smpl'] = smpl_output.vertices[
0].cpu().numpy() # SMPL vertex in original smpl space
pred_output[
'pred_vertices_img'] = pred_vertices_img # SMPL vertex in image space
pred_output[
'pred_joints_img'] = pred_joints_vis_img # SMPL joints in image space
pred_rotmat_tensor = torch.zeros((1, 24, 3, 4),
dtype=torch.float32)
pred_rotmat_tensor[:, :, :, :3] = pred_rotmat.detach().cpu()
pred_aa_tensor = gu.rotation_matrix_to_angle_axis(
pred_rotmat_tensor.squeeze())
pred_output['pred_body_pose'] = pred_aa_tensor.cpu().numpy(
).reshape(1, 72)
pred_output['pred_rotmat'] = pred_rotmat.detach().cpu().numpy(
) # (1, 24, 3, 3)
pred_output['pred_betas'] = pred_betas.detach().cpu().numpy(
) # (1, 10)
pred_output['pred_camera'] = pred_camera
pred_output['bbox_top_left'] = bboxTopLeft
pred_output['bbox_scale_ratio'] = boxScale_o2n
pred_output['faces'] = self.smpl.faces
if self.use_smplx:
img_center = np.array(
(img_original.shape[1], img_original.shape[0])) * 0.5
# right hand
pred_joints = smpl_output.right_hand_joints[0].cpu().numpy(
)
pred_joints_bbox = convert_smpl_to_bbox(
pred_joints, camScale, camTrans)
pred_joints_img = convert_bbox_to_oriIm(
pred_joints_bbox, boxScale_o2n, bboxTopLeft,
img_original.shape[1], img_original.shape[0])
pred_output[
'right_hand_joints_img_coord'] = pred_joints_img
# left hand
pred_joints = smpl_output.left_hand_joints[0].cpu().numpy()
pred_joints_bbox = convert_smpl_to_bbox(
pred_joints, camScale, camTrans)
pred_joints_img = convert_bbox_to_oriIm(
pred_joints_bbox, boxScale_o2n, bboxTopLeft,
img_original.shape[1], img_original.shape[0])
pred_output['left_hand_joints_img_coord'] = pred_joints_img
pred_output_list.append(pred_output)
return pred_output_list
def intergration_copy_paste(pred_body_list, pred_hand_list, smplx_model,
image_shape):
integral_output_list = list()
for i in range(len(pred_body_list)):
body_info = pred_body_list[i]
hand_info = pred_hand_list[i]
if body_info is None or hand_info is None:
integral_output_list.append(None)
continue
# copy and paste
pred_betas = torch.from_numpy(body_info['pred_betas']).cuda()
pred_rotmat = torch.from_numpy(body_info['pred_rotmat']).cuda()
if hand_info['right_hand'] is not None:
right_hand_pose = torch.from_numpy(
hand_info['right_hand']['pred_hand_pose'][:, 3:]).cuda()
right_hand_global_orient = torch.from_numpy(
hand_info['right_hand']['pred_hand_pose'][:, :3]).cuda()
right_hand_local_orient = transfer_hand_wrist(
smplx_model, pred_rotmat[0], right_hand_global_orient,
'right_hand', 'l2g')
pred_rotmat[0, 21] = right_hand_local_orient
else:
right_hand_pose = torch.from_numpy(
np.zeros((1, 45), dtype=np.float32)).cuda()
if hand_info['left_hand'] is not None:
left_hand_pose = torch.from_numpy(
hand_info['left_hand']['pred_hand_pose'][:, 3:]).cuda()
left_hand_global_orient = torch.from_numpy(
hand_info['left_hand']['pred_hand_pose'][:, :3]).cuda()
left_hand_local_orient = transfer_hand_wrist(
smplx_model, pred_rotmat[0], left_hand_global_orient,
'left_hand', 'l2g')
pred_rotmat[0, 20] = left_hand_local_orient
else:
left_hand_pose = torch.from_numpy(
np.zeros((1, 45), dtype=np.float32)).cuda()
# smplx_output = smplx_model(
# betas = pred_betas,
# body_pose = pred_rotmat[:,1:],
# global_orient = pred_rotmat[:,0].unsqueeze(1),
# right_hand_pose = right_hand_pose,
# left_hand_pose= left_hand_pose,
# pose2rot = False)
#Convert rot_mat to aa since hands are always in aa
pred_aa = rotmat3x3_to_angleaxis(pred_rotmat)
pred_aa = pred_aa.view(pred_aa.shape[0], -1)
smplx_output = smplx_model(betas=pred_betas,
body_pose=pred_aa[:, 3:],
global_orient=pred_aa[:, :3],
right_hand_pose=right_hand_pose,
left_hand_pose=left_hand_pose,
pose2rot=True)
pred_vertices = smplx_output.vertices
pred_vertices = pred_vertices[0].detach().cpu().numpy()
pred_joints_3d = smplx_output.joints
pred_joints_3d = pred_joints_3d[0].detach().cpu().numpy()
camScale = body_info['pred_camera'][0]
camTrans = body_info['pred_camera'][1:]
bbox_top_left = body_info['bbox_top_left']
bbox_scale_ratio = body_info['bbox_scale_ratio']
integral_output = dict()
integral_output['pred_vertices_smpl'] = pred_vertices
integral_output['faces'] = smplx_model.faces
integral_output['bbox_scale_ratio'] = bbox_scale_ratio
integral_output['bbox_top_left'] = bbox_top_left
integral_output['pred_camera'] = body_info['pred_camera']
pred_rotmat_tensor = torch.zeros((1, 24, 3, 4), dtype=torch.float32)
pred_rotmat_tensor[:, :, :, :3] = pred_rotmat.detach().cpu()
pred_aa_tensor = gu.rotation_matrix_to_angle_axis(
pred_rotmat_tensor.squeeze())
integral_output['pred_body_pose'] = pred_aa_tensor.cpu().numpy(
).reshape(1, 72)
integral_output['pred_betas'] = pred_betas.detach().cpu().numpy()
integral_output['pred_left_hand_pose'] = left_hand_pose.detach().cpu(
).numpy()
integral_output['pred_right_hand_pose'] = right_hand_pose.detach().cpu(
).numpy()
# convert mesh to original image space (X,Y are aligned to image)
pred_vertices_bbox = convert_smpl_to_bbox(pred_vertices, camScale,
camTrans)
pred_vertices_img = convert_bbox_to_oriIm(pred_vertices_bbox,
bbox_scale_ratio,
bbox_top_left,
image_shape[1],
image_shape[0])
integral_output['pred_vertices_img'] = pred_vertices_img
integral_output_list.append(integral_output)
return integral_output_list
def body_regress(self, img_original, body_bbox_list):
"""
args:
img_original: original raw image (BGR order by using cv2.imread)
body_bbox: bounding box around the target: (minX, minY, width, height)
outputs:
pred_vertices_img:
pred_joints_vis_img:
pred_rotmat
pred_betas
pred_camera
bbox: [bbr[0], bbr[1],bbr[0]+bbr[2], bbr[1]+bbr[3]])
bboxTopLeft: bbox top left (redundant)
boxScale_o2n: bbox scaling factor (redundant)
"""
pred_output_list = list()
for body_bbox in body_bbox_list:
img, norm_img, boxScale_o2n, bboxTopLeft, bbox = process_image_bbox(
img_original, body_bbox, input_res=constants.IMG_RES)
bboxTopLeft = np.array(bboxTopLeft)
# bboxTopLeft = bbox['bboxXYWH'][:2]
if img is None:
pred_output_list.append(None)
continue
with torch.no_grad():
# model forward
pred_rotmat, pred_betas, pred_camera = self.model_regressor(norm_img.to(self.device))
#Convert rot_mat to aa since hands are always in aa
# pred_aa = rotmat3x3_to_angle_axis(pred_rotmat)
pred_aa = gu.rotation_matrix_to_angle_axis(pred_rotmat).cuda()
pred_aa = pred_aa.reshape(pred_aa.shape[0], 72)
smpl_output = self.smpl(
betas=pred_betas,
body_pose=pred_aa[:,3:],
global_orient=pred_aa[:,:3],
pose2rot=True)
pred_vertices = smpl_output.vertices
pred_joints_3d = smpl_output.joints
pred_vertices = pred_vertices[0].cpu().numpy()
pred_camera = pred_camera.cpu().numpy().ravel()
camScale = pred_camera[0] # *1.15
camTrans = pred_camera[1:]
pred_output = dict()
# Convert mesh to original image space (X,Y are aligned to image)
# 1. SMPL -> 2D bbox
# 2. 2D bbox -> original 2D image
pred_vertices_bbox = convert_smpl_to_bbox(pred_vertices, camScale, camTrans)
pred_vertices_img = convert_bbox_to_oriIm(
pred_vertices_bbox, boxScale_o2n, bboxTopLeft, img_original.shape[1], img_original.shape[0])
# Convert joint to original image space (X,Y are aligned to image)
pred_joints_3d = pred_joints_3d[0].cpu().numpy() # (1,49,3)
pred_joints_vis = pred_joints_3d[:,:3] # (49,3)
pred_joints_vis_bbox = convert_smpl_to_bbox(pred_joints_vis, camScale, camTrans)
pred_joints_vis_img = convert_bbox_to_oriIm(
pred_joints_vis_bbox, boxScale_o2n, bboxTopLeft, img_original.shape[1], img_original.shape[0])
# Output
pred_output['img_cropped'] = img[:, :, ::-1]
pred_output['pred_vertices_smpl'] = smpl_output.vertices[0].cpu().numpy() # SMPL vertex in original smpl space
pred_output['pred_vertices_img'] = pred_vertices_img # SMPL vertex in image space
pred_output['pred_joints_img'] = pred_joints_vis_img # SMPL joints in image space
pred_aa_tensor = gu.rotation_matrix_to_angle_axis(pred_rotmat.detach().cpu()[0])
pred_output['pred_body_pose'] = pred_aa_tensor.cpu().numpy().reshape(1, 72)
pred_body_pose = pred_output['pred_body_pose']
pred_output['pred_rotmat'] = pred_rotmat.detach().cpu().numpy() # (1, 24, 3, 3)
pred_output['pred_betas'] = pred_betas.detach().cpu().numpy() # (1, 10)
pred_output['pred_camera'] = pred_camera
pred_output['bbox_top_left'] = bboxTopLeft
pred_output['bbox_scale_ratio'] = boxScale_o2n
pred_output['faces'] = self.smpl.faces
if self.use_smplx:
img_center = np.array((img_original.shape[1], img_original.shape[0]) ) * 0.5
# right hand
pred_joints = smpl_output.right_hand_joints[0].cpu().numpy()
pred_joints_bbox = convert_smpl_to_bbox(pred_joints, camScale, camTrans)
pred_joints_img = convert_bbox_to_oriIm(
pred_joints_bbox, boxScale_o2n, bboxTopLeft, img_original.shape[1], img_original.shape[0])
pred_output['right_hand_joints_img_coord'] = pred_joints_img
# left hand
pred_joints = smpl_output.left_hand_joints[0].cpu().numpy()
pred_joints_bbox = convert_smpl_to_bbox(pred_joints, camScale, camTrans)
pred_joints_img = convert_bbox_to_oriIm(
pred_joints_bbox, boxScale_o2n, bboxTopLeft, img_original.shape[1], img_original.shape[0])
pred_output['left_hand_joints_img_coord'] = pred_joints_img
pred_output_list.append(pred_output)
if self.body_only is True:
with open('/home/student/bodyonly_side/' + 'results.txt', 'a') as file_handle:
file_handle.write('\nframe_id:')
file_handle.write(str(self.frame_id))
file_handle.write('\npred_body_pose:')
file_handle.write(str(pred_body_pose))
file_handle.write('\npred_joints_vis_img:')
file_handle.write(str(pred_joints_vis_img))
#save images(only body_module has visualizer)
# extract mesh for rendering (vertices in image space and faces) from pred_output_list
pred_mesh_list = demo_utils.extract_mesh_from_output(pred_output_list)
# visualization
res_img = self.visualizer.visualize(img_original,pred_mesh_list = pred_mesh_list, body_bbox_list = body_bbox_list)
# show result in the screen
res_img = res_img.astype(np.uint8)
# ImShow(res_img)
cv2.imwrite('/home/student/bodyonly_side/' +str(self.frame_id)+'.jpg',res_img)
return pred_output_list
