def meta_imitate(opt, imitator, prior_tgt_path, save_imgs=True, visualizer=None):
src_path = opt.src_path
all_tgt_paths = scan_tgt_paths(prior_tgt_path, itv=40)
output_dir = opt.output_dir
out_img_dir, out_pair_dir = mkdirs([os.path.join(output_dir, 'imgs'), os.path.join(output_dir, 'pairs')])
img_pair_list = []
for t in tqdm(range(len(all_tgt_paths))):
tgt_path = all_tgt_paths[t]
preds = imitator.inference([tgt_path], visualizer=visualizer, cam_strategy=opt.cam_strategy, verbose=False)
tgt_name = os.path.split(tgt_path)[-1]
out_path = os.path.join(out_img_dir, 'pred_' + tgt_name)
if save_imgs:
cv_utils.save_cv2_img(preds[0], out_path, normalize=True)
write_pair_info(imitator.src_info, imitator.tsf_info,
os.path.join(out_pair_dir, '{:0>8}.pkl'.format(t)), imitator=imitator,
only_vis=opt.only_vis)
img_pair_list.append((src_path, tgt_path))
if save_imgs:
write_pickle_file(os.path.join(output_dir, 'pairs_meta.pkl'), img_pair_list)
def write_pair_info(src_info, tsf_info, out_file, imitator, only_vis):
"""
Args:
src_info:
tsf_info:
out_file:
imitator:
Returns:
"""
pair_data = dict()
pair_data['from_face_index_map'] = src_info['fim'][0][:, :, None].cpu().numpy()
pair_data['to_face_index_map'] = tsf_info['fim'][0][:, :, None].cpu().numpy()
pair_data['T'] = tsf_info['T'][0].cpu().numpy()
pair_data['warp'] = tsf_info['tsf_img'][0].cpu().numpy()
pair_data['smpls'] = torch.cat([src_info['theta'], tsf_info['theta']], dim=0).cpu().numpy()
pair_data['j2d'] = torch.cat([src_info['j2d'], tsf_info['j2d']], dim=0).cpu().numpy()
tsf_f2verts, tsf_fim, tsf_wim = imitator.render.render_fim_wim(tsf_info['cam'], tsf_info['verts'])
tsf_p2verts = tsf_f2verts[:, :, :, 0:2]
tsf_p2verts[:, :, :, 1] *= -1
T_cycle = imitator.render.cal_bc_transform(tsf_p2verts, src_info['fim'], src_info['wim'])
pair_data['T_cycle'] = T_cycle[0].cpu().numpy()
# back_face_ids = mesh.get_part_face_ids(part_type='head_back')
# tsf_p2verts[:, back_face_ids] = -2
# T_cycle_vis = imitator.render.cal_bc_transform(tsf_p2verts, src_info['fim'], src_info['wim'])
# pair_data['T_cycle_vis'] = T_cycle_vis[0].cpu().numpy()
# for key, val in pair_data.items():
# print(key, val.shape)
write_pickle_file(out_file, pair_data)
if len(hull.vertices) == 3:
pt_normal[idx] = f_n * (-1)
return pt_position, pt_normal
if __name__ == '__main__':
data_root = 'data/Future-3D-Reconstruction'
model_dir = os.path.join(data_root, 'train', 'model')
verts_normals_dir = os.path.join(data_root, 'train', 'verts_normals')
if not os.path.exists(verts_normals_dir):
os.mkdir(verts_normals_dir)
model_IDs = os.listdir(model_dir)
model_IDs = [ID for ID in model_IDs if ID[-4:] == '.obj']
print('There are {} models'.format(len(model_IDs)))
widgets = ['Progress: ', Percentage(), ' ', Bar('#'), ' ', Timer(), ' ', ETA()]
total = len(model_IDs)
pbar = ProgressBar(widgets=widgets, maxval=10*total).start()
for i, model_ID in enumerate(model_IDs):
model_path = os.path.join(model_dir, model_ID)
verts_normals_path = os.path.join(verts_normals_dir, model_ID.replace('.obj', '.pth'))
face_pts, face_axis = readFaceInfo(model_path)
#print(face_pts.shape, face_axis.shape)
verts = get_obj_vertex_ali(model_path)
#print(verts.shape)
verts, normals = generate_normal(verts, face_pts, face_axis)
#print(verts.shape, normals.shape)
write_pickle_file(verts_normals_path, {'verts': verts, 'normals': normals})
pbar.update(10 * i + 1)
pbar.finish()
if __name__ == '__main__':
data_root = 'data/Future-3D-Reconstruction'
camera_dir = os.path.join(data_root, 'train', 'cameras')
with open(os.path.join(data_root, 'train', 'data_info',
'train_set.json')) as f:
train_set_info = json.load(f)
total = len(train_set_info)
for i in range(total):
model_ID = train_set_info[i]['model']
img = Image.open(
os.path.join(data_root, 'train', 'image',
train_set_info[i]['image'])).convert('RGB')
trans = np.array(train_set_info[i]['pose']['translation'])
rot = np.array(train_set_info[i]['pose']['rotation'])
init_blender(img.size)
try:
#fov = math.radians(train_set_info[i]['fov'])
fov = train_set_info[i]['fov']
cam = add_camera((0, 0, 0), fov, 'camera')
K_blender = get_calibration_matrix_K_from_blender(cam.data)
K = np.array(K_blender)
clear_scene()
clear_mv()
except:
continue
cam_path = os.path.join(camera_dir, model_ID + '.pkl')
write_pickle_file(cam_path, {'K': K})
print('[ {} / {} ] Saved to {}'.format(i, total, cam_path))
print('Finished!')
