Troubleshooting

Troubleshooting

ModuleNotFoundError

Confirm the virtual environment is active and install the project dependencies:

.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Use requirements-deep.txt when the selected training/evaluation path needs the deep-learning stack.

PowerShell script execution is blocked

For the current terminal process:

Set-ExecutionPolicy -Scope Process -ExecutionPolicy RemoteSigned

MEVA file/annotation mismatch

Use the annotation inventory and manifest-building scripts to verify that referenced media exists locally. Do not silently keep manifest records whose source video is unavailable.

Training stopped mid-run

Use the resumable training workflow rather than starting a new experiment under the same output directory without checking existing checkpoints/history.

Relevant entry point:

scripts/train_meva_activity_resumable.py

Very large repository / push failure

Do not commit .pt, .pth, video files, raw datasets, .venv/ or generated outputs/. Large trained checkpoints belong in controlled artifact storage or a release/model registry if redistribution is appropriate.

git status shows data/

That means the local data directory is not necessarily ignored at the directory root. Add the appropriate local data path to .gitignore without removing dataset-preparation code or small redistributable metadata that the project intentionally tracks.

Wiki push fails with repository not found

GitHub creates the separate Wiki Git repository after an initial Wiki page exists. Create/save the first Wiki page in the GitHub UI once, then rerun the documentation publisher.