Training Problems

[ITBoy-China]

Hello, thanks for your great work. I met "ValueError: Surface level must be within volume data range" when I use my own data to train. I knew that occured in validation step , so I turned off validation step and it can be trained normally. But when I extracted mesh and used the trained model. The error came again. I'd appreciate it if you could give me some advices.

Errors like this ↓
And my dataset is indoor and I use train_indoor.yaml as my configuration.

Here is my dataset. And I just removed some detailed files.

/mnt/share_disk/shlzhang/data/NeuralReconW/right_camera/scene
|-- cache_sgs
| -- splits | |-- rays1_meta_info.json | |-- rgbs1_meta_info.json | |-- split_0 | | |-- rays1.h5 | | -- rgbs1.h5
......
|-- config.yaml
|-- dense
| |-- images
| | |-- 1663582290115993.jpg
| | |-- 1663582290215977.jpg
| | |-- 1663582290315950.jpg
......
| -- sparse | |-- cameras.bin | |-- images.bin | -- points3D.bin
|-- segmentation_vis
| |-- 1663582290115993.png
| |-- 1663582290215977.png
| |-- 1663582290315950.png
......
|-- semantic_maps
| |-- 1663582290115993.npz
| |-- 1663582290215977.npz
| |-- 1663582290315950.npz
......
|-- split_0.tsv
`-- trash_images

[Burningdust21]

Hi, the error during extracting mesh means no valid surface is reconstructed. You should check: 1) the bounding box is set correctly; 2) the SFM point cloud is not too sparse; 3) If your camera is within sampling range, i.e. an indoor scene, use the indoor config.

If the problem still exists, you can post the tensorborad log here or send me your data, I'm happy to take a look.