# parser.add_argument("depth_image_filename", type=str, help="Sequence of the dataset")
args = parser.parse_args()
args.ho3d_path = '/home/tpatten/v4rtemp/datasets/HandTracking/HO3D_v2'
args.mask_dir = '/home/tpatten/Data/Hands/HO3D_V2/HO3D_v2_segmentations_rendered/'
args.scene = 'ABF10'
args.model_render = False
ren = None
if args.model_render:
width = 640
height = 480
renderer_modalities = []
renderer_modalities.append('rgb')
renderer_modalities.append('depth')
renderer_mode = '+'.join(renderer_modalities)
ren = renderer.create_renderer(width, height, 'python', mode=renderer_mode, shading='flat')
add_objects_to_renderer(ren)
frame_ids = sorted(os.listdir(os.path.join(args.ho3d_path, data_split, args.scene, 'rgb')))
for i in range(500, len(frame_ids)):
# Get the id
frame_id = frame_ids[i].split('.')[0]
# Load the image
depth, mask_obj, mask_hand, mask_scene = load_data(args.ho3d_path, args.mask_dir, args.scene, frame_id)
# Load the object pose
obj_id, object_mesh, object_pose, camK = load_object_and_pose(args.ho3d_path, args.scene, frame_id)
if args.model_render:
rendered_depth = render_depth(ren, obj_id, object_pose, camK)
dp_model['models_info_path'], keys_to_int=True)
# Get sets of symmetry transformations for the object models.
models_sym = None
if p['error_type'] in ['mssd', 'mspd']:
models_sym = {}
for obj_id in dp_model['obj_ids']:
models_sym[obj_id] = misc.get_symmetry_transformations(
models_info[obj_id], p['max_sym_disc_step'])
# Initialize a renderer.
ren = None
if p['error_type'] in ['vsd', 'cus']:
misc.log('Initializing renderer...')
width, height = dp_split['im_size']
ren = renderer.create_renderer(
width, height, p['renderer_type'], mode='depth')
for obj_id in dp_model['obj_ids']:
ren.add_object(obj_id, dp_model['model_tpath'].format(obj_id=obj_id))
# Load the estimation targets.
targets = inout.load_json(
os.path.join(dp_split['base_path'], p['targets_filename']))
# Organize the targets by scene, image and object.
misc.log('Organizing estimation targets...')
targets_org = {}
for target in targets:
targets_org.setdefault(target['scene_id'], {}).setdefault(
target['im_id'], {})[target['obj_id']] = target
# Load pose estimates.
# Image size and K for the RGB image (potentially with SSAA). im_size_rgb = [int(round(x * float(ssaa_fact))) for x in dp_camera['im_size']] K_rgb = dp_camera['K'] * ssaa_fact # Intrinsic parameters for RGB rendering. fx_rgb, fy_rgb, cx_rgb, cy_rgb =\ K_rgb[0, 0], K_rgb[1, 1], K_rgb[0, 2], K_rgb[1, 2] # Intrinsic parameters for depth rendering. K = dp_camera['K'] fx_d, fy_d, cx_d, cy_d = K[0, 0], K[1, 1], K[0, 2], K[1, 2] # Create RGB and depth renderers (two are created because the RGB has a higher # resolution if SSAA is used). width_rgb, height_rgb = im_size_rgb[0], im_size_rgb[1] ren_rgb = renderer.create_renderer( width_rgb, height_rgb, renderer_type, mode='rgb', shading=shading) ren_rgb.set_light_ambient_weight(ambient_weight) width_depth, height_depth, = dp_camera['im_size'][0], dp_camera['im_size'][1] ren_depth = renderer.create_renderer( width_depth, height_depth, renderer_type, mode='depth') # Render training images for all object models. for obj_id in obj_ids: # Add the current object model to the renderer. ren_rgb.add_object(obj_id, dp_model['model_tpath'].format(obj_id=obj_id)) ren_depth.add_object(obj_id, dp_model['model_tpath'].format(obj_id=obj_id)) # Prepare output folders. misc.ensure_dir(os.path.dirname(out_rgb_tpath.format(
if p['scene_ids']:
scene_ids_curr = set(scene_ids_curr).intersection(p['scene_ids'])
# Rendering mode.
renderer_modalities = []
if p['vis_rgb']:
renderer_modalities.append('rgb')
if p['vis_depth_diff'] or (p['vis_rgb'] and p['vis_rgb_resolve_visib']):
renderer_modalities.append('depth')
renderer_mode = '+'.join(renderer_modalities)
# Create a renderer.
width, height = dp_split['im_size']
ren = renderer.create_renderer(width,
height,
p['renderer_type'],
mode=renderer_mode,
shading='flat')
# Load object models.
models = {}
for obj_id in dp_model['obj_ids']:
misc.log('Loading 3D model of object {}...'.format(obj_id))
model_path = dp_model['model_tpath'].format(obj_id=obj_id)
model_color = None
if not p['vis_orig_color']:
model_color = tuple(colors[(obj_id - 1) % len(colors)])
ren.add_object(obj_id, model_path, surf_color=model_color)
scene_ids = dataset_params.get_present_scene_ids(dp_split)
for scene_id in scene_ids: