How to fix "Cannot find valid data directory for uid" error?

All runs in virtual box Ubuntu linux

[train_metadata.json](https://github.com/3DTopia/OpenLRM/files/14841706/train_metadata.json)
[val_metadata.json](https://github.com/3DTopia/OpenLRM/files/14841707/val_metadata.json)
There is jsons with links to my dataset of png files . When I tried to start training model I have "Cannot find valid data directory for uid" error . It seems that trouble can be in links or config files that is provided downwards but I`m not sure

root_dirs:
./DATA_set/
meta_path:
train: ./DATA_set/train_metadata.json
val: ./DATA_set/val_metadata.json

Maybe there are some naming rules or something related to image tracking?

Start by checking the provided train_metadata.json and val_metadata.json together with the root_dirs and meta_path settings in the training configuration. Reproduce the error in the reported Ubuntu VirtualBox setup and trace how the training entry point resolves each UID to a data directory. Done means the dataset paths and naming requirements are identified and training no longer reports that no valid data directory exists.

Written by the indexing model from the issue text.