# Load models of objects that appear in the current scene
obj_ids = set([gt['obj_id'] for gts in scene_gt.values() for gt in gts])
models = {}
for obj_id in obj_ids:
models[obj_id] = inout.load_ply(par['model_mpath'].format(obj_id))
# Visualize GT poses in the selected images
im_ids_curr = sorted(scene_info.keys())
if im_ids:
im_ids_curr = set(im_ids_curr).intersection(im_ids)
for im_id in im_ids_curr:
print('scene: {}, im: {}'.format(scene_id, im_id))
# Load the images
rgb = inout.read_im(par['test_rgb_mpath'].format(scene_id, im_id))
depth = inout.read_depth(par['test_depth_mpath'].format(
scene_id, im_id))
depth = depth.astype(np.float) # [mm]
depth *= par['cam']['depth_scale'] # to [mm]
# Render the objects at the ground truth poses
im_size = (depth.shape[1], depth.shape[0])
ren_rgb = np.zeros(rgb.shape, np.float)
ren_depth = np.zeros(depth.shape, np.float)
gt_ids_curr = range(len(scene_gt[im_id]))
if gt_ids:
gt_ids_curr = set(gt_ids_curr).intersection(gt_ids)
for gt_id in gt_ids_curr:
gt = scene_gt[im_id][gt_id]
model = models[gt['obj_id']]
for res_path in res_paths:
#t = time.time()
# Parse image ID and object ID from the file name
filename = os.path.basename(res_path).split('.')[0]
im_id_prev = im_id
im_id, obj_id = map(int, filename.split('_'))
print('Calculating {} error - scene: {}, im: {}, obj: {}'.format(
error_type, scene_id, im_id, obj_id))
# Load depth image if VSD is selected
if error_type == 'vsd':
if im_id != im_id_prev:
depth_path = dp['test_depth_mpath'].format(scene_id, im_id)
depth_im = inout.read_depth(depth_path)
depth_im *= dp['cam']['depth_scale'] # to [mm]
# Load camera matrix
if error_type in ['vsd', 'cou']:
K = scene_info[im_id]['cam_K']
# Load pose estimates
ests = inout.load_poses(res_path)
# Sort the estimates by score (in descending order)
ests_sorted = sorted(enumerate(ests),
key=lambda x: x[1]['score'],
reverse=True)
# Consider only the top N estimated poses
obj_id = scene_id # The object id is the same as scene id for this dataset
model = inout.load_ply(model_mpath.format(obj_id))
# Transformation which was applied to the object models (its inverse will
# be applied to the GT poses):
# 1) Translate the bounding box center to the origin
t_model = bbox_cens[obj_id - 1, :].reshape((3, 1))
im_id_out = 0
for im_id in im_ids:
# if im_id % 10 == 0:
print('scene,view: ' + str(scene_id) + ',' + str(im_id))
# Load the RGB and depth image
rgb = inout.read_im(rgb_in_mpath.format(scene_id, im_id))
depth = inout.read_depth(depth_in_mpath.format(scene_id, im_id))
depth *= 10.0 # Convert depth map to [100um]
# Save the RGB and depth image
inout.write_im(rgb_out_mpath.format(scene_id, im_id_out), rgb)
inout.write_depth(depth_out_mpath.format(scene_id, im_id_out), depth)
scene_info[im_id_out] = {
'cam_K': par['cam']['K'].flatten().tolist()
}
# Process the GT poses
poses = load_tejani_poses(pose_mpath.format(scene_id, im_id))
scene_gt[im_id_out] = []
for pose in poses: