4th place solution (tattaka's part)

Requirements

16GB x 4 VRAM (trained on NVIDIA RTX A4000 x 4).

Environment

Use Kaggle Docker.
Follow tattaka/ml_environment to build the environment.
You can run ./RUN-KAGGLE-GPU-ENV.sh and launch the docker container.

Usage

1. Place competition data in the input directory
2. Run all the input/make_segmentation_label_v0106.ipynb for making grand truth.
3. Run the following script for training.
   • cd src/exp122 && sh train_resnetrs50.sh
4. Run all the src/exp122/export_onnx.ipynb for converting onnx file.
5. Upload to kaggle dataset, and Run all https://www.kaggle.com/code/tattaka/converting-pytorch-checkpoints-to-tensorrt-models

License

MIT