This is a PyTorch implementation of the ECCV2018 paper "Learning to Navigate for Fine-grained Classification" (Ze Yang, Tiange Luo, Dong Wang, Zhiqiang Hu, Jun Gao, Liwei Wang).you can find original repository from github.com/yangze0930/NTS-Net

• python 3+
• pytorch 0.4+
• numpy
• datetime

Download the CUB-200-2011 datasets and put it in the root directory named CUB_200_2011, You can also try other fine-grained datasets.

I recommend taht your dataset should be like as below:

dataset
|---images
|-------image01
|-------image02
|-------....
|---X_train.txt
|---X_test.txt
|---y_train.txt
|---y_test.txt

If you want to train the NTS-Net, just run python train.py. You may need to change the configurations in config.py. The parameter PROPOSAL_NUM is M in the original paper and the parameter CAT_NUM is K in the original paper. During training, the log file and checkpoint file will be saved in save_dir directory. You can change the parameter resume to choose the checkpoint model to resume.

The test.py doesn't work on your own dataset, I will make it soon. And also,I am glad to see that anybody could contribute to this repository. If you want to test the NTS-Net on CUB dataset, just run python test.py. You need to specify the test_model in config.py to choose the checkpoint model for testing.

If you want to train NTS-Net on your own dataset, you need to prepare the following:

1. Divide your data set into two parts, one is the training set and the other is the test set.
2. There are four files in the data folder: X_train.txt, y_train.txt, X_test.txt, y_test.txt

X_train.txt shows as follows

/data/path/to/your/own/data/image01
/data/path/to/your/own/data/image02
/data/path/to/your/own/data/image03
/data/path/to/your/own/data/image04

y_rain.txt shows as follows

1
1
2
2

Your label_map file should be placed in the data folder,

label01,1
label02,2
label03,3
label04,4

3. Modify FC_NUMS in config.py to the total number of classes in your dataset, and DATASET_PATH as the path to your dataset.

4. Before training, please calculate the mean and variance on the dataset. just run

Python core/utils.py

Then modify the MEAN and STD in config.py

5. Confirm your model save location and modify the save_dir field in config.py to adjust the value of the batch_size and dataloader_num_workers parameters for your machine. You can modify it in config.py.

6. modify the own_dataset as True in config.py, run python train.py to start training your own dataset.

note:

If your machine is multi-GPU, you can modify it in train.py Os.environ['CUDA_VISIBLE_DEVICES'] = '0', refer to the official pytorch documentation

the official repository provides a checkpoint model trained by author yangze0930,please check here to get model.If you test on provided model, you will get a 87.6% test accuracy as paper shows.

• add densenet as feature extractor, it will be soon
• edit test.py to test on own trained model and dataset, it will be soon
• add squeezenet and mobilenet as feature extractor once I have enough time

@inproceedings{Yang2018Learning,
author = {Yang, Ze and Luo, Tiange and Wang, Dong and Hu, Zhiqiang and Gao, Jun and Wang, Liwei},
title = {Learning to Navigate for Fine-grained Classification},
booktitle = {ECCV},
year = {2018}
}