dimension misamatch in fut_traj[:fut_traj_scence_centric.shape[0], :] = fut_traj_scence_centric when getting future trajectory info

[HHADS-dev]

Hi, I met a dimension mismatch error when I creating the trajectory info for uniad using

def _get_future_traj_info(nusc, sample, predict_steps=16):
    sample_token = sample['token']
    ann_tokens = np.array(sample['anns'])
    sd_rec = nusc.get('sample', sample_token)
    fut_traj_all = []
    fut_traj_valid_mask_all = []
    _, boxes, _ = nusc.get_sample_data(sd_rec['data']['LIDAR_TOP'], selected_anntokens=ann_tokens)
    predict_helper = PredictHelper(nusc)
    for i, ann_token in enumerate(ann_tokens):
        box = boxes[i]
        instance_token = nusc.get('sample_annotation', ann_token)['instance_token']
        fut_traj_local = predict_helper.get_future_for_agent(instance_token,
                                                             sample_token,
                                                             seconds=predict_steps//2,
                                                             in_agent_frame=True)

        fut_traj = np.zeros((predict_steps, 2))
        fut_traj_valid_mask = np.zeros((predict_steps, 2))
        if fut_traj_local.shape[0] > 0:
            # trans = box.center
            # trans = np.array([0, 0, 0])
            # rot = Quaternion(matrix=box.rotation_matrix)
            # fut_traj_scence_centric = convert_local_coords_to_global(fut_traj_local, trans, rot)  
            fut_traj_scence_centric = fut_traj_local
            print(f"fut_traj.shape: {fut_traj.shape}, fut_traj_scence_centric.shape: {fut_traj_scence_centric.shape}")
            fut_traj[:fut_traj_scence_centric.shape[0], :] = fut_traj_scence_centric
            fut_traj_valid_mask[:fut_traj_scence_centric.shape[0], :] = 1
        fut_traj_all.append(fut_traj)
        fut_traj_valid_mask_all.append(fut_traj_valid_mask)
    if len(ann_tokens) > 0:
        fut_traj_all = np.stack(fut_traj_all, axis=0)
        fut_traj_valid_mask_all = np.stack(fut_traj_valid_mask_all, axis=0)
    else:
        fut_traj_all = np.zeros((0, predict_steps, 2))
        fut_traj_valid_mask_all = np.zeros((0, predict_steps, 2))
    return fut_traj_all, fut_traj_valid_mask_all

the error shows that

Traceback (most recent call last):
  File "tools/create_data.py", line 85, in <module>
    nuscenes_data_prep(
  File "tools/create_data.py", line 28, in nuscenes_data_prep
    nuscenes_converter.create_nuscenes_infos(
  File "/mnt/ws-frb/users/yiliuhh/mmpretraining/ViDAR/UniAD/tools/data_converter/uniad_nuscenes_converter.py", line 86, in create_nuscenes_infos
    train_nusc_infos, val_nusc_infos = _fill_trainval_infos(
  File "/mnt/ws-frb/users/yiliuhh/mmpretraining/ViDAR/UniAD/tools/data_converter/uniad_nuscenes_converter.py", line 321, in _fill_trainval_infos
    future_traj_all, future_traj_valid_mask_all = _get_future_traj_info(nusc, sample)
  File "/mnt/ws-frb/users/yiliuhh/mmpretraining/ViDAR/UniAD/tools/data_converter/uniad_nuscenes_converter.py", line 199, in _get_future_traj_info
    fut_traj[:fut_traj_scence_centric.shape[0], :] = fut_traj_scence_centric
ValueError: could not broadcast input array from shape (17,2) into shape (16,2)

it seems like there is a dimension mismatch between fut_traj.shape and fut_traj_scence_centric fut_traj.shape: (16, 2), fut_traj_scence_centric.shape: (17, 2) have anyone met similar issue before? are there any suggestions/ thanks!

[HHADS-dev]

the value of the fut_traj_local is

array([[ 0.00000000e+00,  0.00000000e+00],
       [ 0.00000000e+00,  0.00000000e+00],
       [ 0.00000000e+00,  0.00000000e+00],
       [ 0.00000000e+00,  0.00000000e+00],
       [-3.86562523e-02, -2.90808211e-02],
       [-3.86562523e-02, -2.90808211e-02],
       [-1.20419962e-01, -1.22078928e-02],
       [-1.95717329e-01, -6.61265921e-03],
       [-3.00412895e-01,  3.03866304e-04],
       [-3.73306838e-01,  5.86174417e-02],
       [-4.46326270e-01,  1.17923112e-01],
       [-4.46326270e-01,  1.17923112e-01],
       [-4.83386618e-01,  1.00131803e-01],
       [-4.96648502e-01,  9.34144811e-02],
       [-5.47519961e-01,  1.05123227e-01],
       [-5.68938635e-01,  1.62892080e-01],
       [-5.91548608e-01,  1.02577992e-01]])

[zzh-yun]

you use your custom data? you need to change the predict_steps=16 to a big number (more than 17)