Question about the MegaDepth dataset for training.

[LuoXubo]

Hi. I'm very interested in your work, XFeat. However, I met some problems when I tried to train the model. I've made the training megadepth dataset according to the instructions in your repository. And the dataset structure is as follows:

    ├── megadepth_root_path
    │   ├── train_data
    │   │   ├── megadepth_indeices
    │   │   │   ├── scene_info_0.1_0.7
    │   │   │   │   ├── 0000_0.1_0.3.npz
    │   │   │   │   ├── ...
    │   │   │   ├── ...
    │   ├── MegaDepth_v1
    │   │   ├── 000
    │   │   │   ├── dense0
    │   │   │   │   ├── depths
    │   │   │   │   ├── images
    │   │   │   ├── dense1
    │   │   ├── ...

The loading part of the dataset in the training code is as follows:

TRAIN_BASE_PATH = f"{config['megadepth_root_path']}/train_data/megadepth_indices"
TRAINVAL_DATA_SOURCE = f"{config['megadepth_root_path']}/MegaDepth_v1"

TRAIN_NPZ_ROOT = f"{TRAIN_BASE_PATH}/scene_info_0.1_0.7"

npz_paths = glob.glob(TRAIN_NPZ_ROOT + '/*.npz')[:]

data = torch.utils.data.ConcatDataset( [MegaDepthDataset(root_dir = TRAINVAL_DATA_SOURCE,
    npz_path = path) for path in tqdm.tqdm(npz_paths, desc="[MegaDepth] Loading metadata")] )

And I met the error when I ran the training code:

Traceback (most recent call last):
  File "train.py", line 279, in <module>
    trainer = Trainer(config_path = '../config.yaml')
  File "train.py", line 39, in __init__
    self.load_config(config_path)
  File "train.py", line 82, in load_config
    data = torch.utils.data.ConcatDataset( [MegaDepthDataset(root_dir = TRAINVAL_DATA_SOURCE,
  File "/home/anaconda3/envs/tavins/lib/python3.8/site-packages/torch/utils/data/dataset.py", line 398, in __init__
    assert len(self.datasets) > 0, 'datasets should not be an empty iterable'  # type: ignore[arg-type]
AssertionError: datasets should not be an empty iterable

I think there must be some problem with the dataset structure. Could you help me with the problem? Or could you please give me an example of the structure of the dataset? Thank you very much.

Looking forward to your reply!

[guipotje]

Hi @LuoXubo,

Thank you for your interest in our work.

The dataset structure seems correct; however, I noticed a typo, "megadepth_indeices," which could be causing an error. Were you able to resolve the issue?

Kind regards.

[LuoXubo]

Hi, thanks for your reply. I've fixed the problem :-)