File "<__array_function__ internals>", line 180, in concatenate ValueError: need at least one array to concatenate

[harrykwon0524]

Hi I am trying to run the file and put down the path for hippocampus dataset, and as a result, I see the error as stated in the title. Is there some additional lines that need to be put? Thank you

Traceback (most recent call last):
File "data_preprocess.py", line 43, in
main()
File "data_preprocess.py", line 39, in main
data = data_obj.pre_process_dataset(cfg)
File "/home/cs2212/Desktop/voxel2mesh-master/data/hippocampus.py", line 97, in pre_process_dataset
inputs_ = np.concatenate(inputs_, axis=0)
File "<array_function internals>", line 180, in concatenate
ValueError: need at least one array to concatenate

[yoterel]

I have the same error.

Further investigation shows that the preprocess step expects .npy files in the dataset folder of Hippocampus/imagesTr, but only .nii.gz files exist (and same goes for imagesTs and labelsTr).
This is using the download link provided in the in the repository (README.md)

[udaranga3001]

It seems I have missed the dataset_init function. I have copied it below and will try to make a commit with the updated hippocampus.py soon. Meantime, you can try the function below. Replace line 75 in hippocampus.py with the function call dataset_init(data_root, multi_stack)

def dataset_init(data_root, multi_stack=None):
    # multi_stack: if image (x) is 4-dim, (i.e. same file has multiple image volumes, you should specify the index of the volume to be used.
    samples = [dir for dir in os.listdir('{}/imagesTr'.format(data_root))]

    inputs = []
    labels = []
    real_sizes = []
    file_names = []

    vals = []
    sizes = [] 

    count = 0
    for itr, sample in enumerate(tqdm(samples)):
        if '.nii.gz' in sample and '._' not in sample and '.npy' not in sample and '.tif' not in sample:
 
            x = nib.load('{}/imagesTr/{}'.format(data_root, sample))
            y = nib.load('{}/labelsTr/{}'.format(data_root, sample)).get_fdata() > 0
            resolution = np.diag(x.header.get_sform())[:3]
            x = x.get_fdata()
            if multi_stack is not None:
                x = x[:, :, :, multi_stack]
            real_size = np.round(np.array(x.shape) * resolution)
            
            file_name = sample
 

            x = torch.from_numpy(x).permute([2, 1, 0]).cuda().float()
            y = torch.from_numpy(y).permute([2, 1, 0]).cuda().float()
            #
            W, H, D = real_size
            W, H, D = int(W), int(H), int(D)
            base_grid = torch.zeros((1, D, H, W, 3))
            w_points = (torch.linspace(-1, 1, W) if W > 1 else torch.Tensor([-1]))
            h_points = (torch.linspace(-1, 1, H) if H > 1 else torch.Tensor([-1])).unsqueeze(-1)
            d_points = (torch.linspace(-1, 1, D) if D > 1 else torch.Tensor([-1])).unsqueeze(-1).unsqueeze(-1)
            base_grid[:, :, :, :, 0] = w_points
            base_grid[:, :, :, :, 1] = h_points
            base_grid[:, :, :, :, 2] = d_points
            grid = base_grid.cuda()
 

            x = F.grid_sample(x[None, None], grid, mode='bilinear', padding_mode='border')[0, 0].cpu().numpy()
            y = F.grid_sample(y[None, None], grid, mode='nearest', padding_mode='border')[0, 0].long().cpu().numpy()


            x = (x - np.mean(x))/np.std(x) 
            np.save('{}/imagesTr/p{}'.format(data_root, file_name), x)
            np.save('{}/labelsTr/p{}'.format(data_root, file_name), y)

            count += 1 

[harrykwon0524]

thank you for the update