rkorv/MCS2022.CarVerification

MCS2022.CarVerification

Competition: https://ods.ai/competitions/mcs_car_verification

#	ROC AUC	Domain 1	Domain 2	Domain 3	Domain 4	Domain 5
1	0.97571	0.985	0.944	0.992	0.982	0.994

Train and eval

Preparation

1. Prepare datasets folder

├── datasets
│   ├── public_test
│   └── vehicle_models
└── MCS2022.CarVerification
    ├── <all_sources>
    ├── checkpoints
    ├── database
    └── logs

2. Install requirements.txt (tested for python3.6):

pip install -r requirements.txt

Train

cd MCS2022.CarVerification
python train.py

Generate database

Change PRETRAIN variable inside generate_database.py

cd MCS2022.CarVerification
python generate_database.py

Inference

Use scripts/ckpt_to_model.py to convert checkpoints to model weights.

Change filenames inside make_submit.py

cd MCS2022.CarVerification
python make_submit.py

References

• https://github.com/layumi/Person_reID_baseline_pytorch
• https://github.com/KevinMusgrave/pytorch-metric-learning
• https://github.com/abhuse/pytorch-efficientnet