R^2-Trans

This is a project website of the paper "R^2-Trans: Fine-Grained Visual Categorization with Redundancy Reduction" This website includes the following materials for testing and checking our results reported in our paper

1. The trained model
2. The test scripts
3. More visualization of our experiments [TODO]

Usage

1. Prepare the environment

We trained and tested our code in the following environments with one A6000 GPU:
ubuntu18.04
python3.7
CUDA 11.1
cuDNN 8.0.5
Pytorch 1.8.1
...

Then install the required packages:

pip install -r requirements.txt

2. Preparing Dataset and Model

We provide trained models (Dropbox) on three different datasets: cub, dogs and nabirds. You can download model files as well as corresponding datasets, and organize them as follows:

.
├── checkpoint
│   ├── cub_R2-Trans_checkpoint.pth
│   ├── dogs_R2-Trans_checkpoint.pth
│   └── nabirds_R2-Trans_checkpoint.pth
├── data
│   ├── cub/
│   ├── dogs/
│   └── nabirds/
└── ···

3. Running

example for cub:

python3 test.py --dataset cub 
                --data_root ./data/CUB_dataset/CUB_200_2011/ 
                --checkpoint ./checkpoint/cub_R2-Trans_checkpoint.pth
                --eval_batch_size 5
                --gpu_ids 0