-
Notifications
You must be signed in to change notification settings - Fork 42
Description
Dear authors,
Thank you for your excellent work on GeoWizard.
I would like to train the model using my own custom dataset. To correctly format my data, could you please share the expected directory structure for the dataset (especially how RGB images, depth, normals, and annotations should be organized)?
Based on your current implementation and Replica dataset usage, I have inferred a possible folder structure as follows. Could you kindly confirm if this is correct or suggest modifications?
<dataset_root>/
└── replica/
├── annotations/
│ └── annos_all.json # Index file with metadata paths
└── results/
├── 00000.pkl # Per-frame metadata (paths, intrinsics, pose, etc.)
├── 00000_rgb.png # RGB image
├── 00000_depth.png # Depth image
├── 00000_normal.png # Normal image
└── ... # More frames
Each .pkl file could contain the paths to the corresponding RGB, depth, and normal files, along with intrinsics and pose data.
It would be extremely helpful if you could confirm whether this format is correct for training, or if there are any additional requirements.
Thank you very much for your support!
Best regards!