- Jiří Stehlík, xstehl23@vutbr.cz
- Filip Černý, xcerny81@vutbr.cz
- Marek Halamka, xhalam16@vutbr.cz
Project dependencies are managed by poetry. To install all dependencies run:
petry installNote
Poetry is unable to install correct PyTorch version for you. If you want to install specific PyTorch version manually (as system package for example), you can install only other dependencies (without torch) via poetry install --without system-deps.
Training scripts are in train/.
Usage:
python train.py <checkpoints_path> <model_name/checkpoint_path> [--fit/--all]Will train model_name on VeRi dataset for at least 15 epochs with batch_size=200.
Then it will be stopped by early stopping based on mAP, min_delta=0.003 and patience=5.
Top 3 checkpoints will be saved to checkpoints_path.
By default, model is trained only on VeRi dataset. If --fit is specified, training is done on FIT dataset only and if --all, training is done on both.
Testing scripts are in train/.
Usage:
python test.py <checkpoint_path>This will evaluate model on both VeRi and FIT datasets.
For model training, you have to obtain and your own copy for datasets. To get more information, look into datasets folder.