NTS-Net-Keras

This project is a tool to build NTS-Net models, written in Keras.

original paper:Learning to Navigate for Fine-grained Classification.

Get CUB-200-2011

Multi-gpu training is supported

Only support tensorflow as backend

Quickstart

Note that currently this project can only be executed in Linux and macOS. You might run into some issues in Windows. Python version: python2.7.

1. Download CUB_200_2011.tgz and extract the tgz file.
2. Install dependencies by running pip install -r requirements.txt.
3. Edit config.py to configure your experiment,you may have to set data_root,num_gpu,batch_size and so on.
4. Run python train.py to train a new model.

Trained model weights

CUDA version

CUDA 9.0 is required

performance

Accuracy on test set is 0.82,which is 5 percent lower than the original implementation.PR is welcomed to help slove this problem.

Acknowledgement

Original implementation NTS-Net,pytorch version.

Author

He Jian ([email protected])

License

MIT