fouriye / cnl-icip2020

Associating Multi-Scale Receptive Fields for Fine-grained Recognition (ICIP2020)

Associating Multi-Scale Receptive Fields for Fine-grained Recognition

By Zihan Ye ([email protected]), Fuyuan Hu, Yin Liu ([email protected]), Zhenping Xia, Fan Lyu ([email protected]) and Pengqing Liu

Introduction

This is a PyTorch implementation for the paper "Associating Multi-Scale Receptive Fields for Fine-grained Recognition" in ICIP2020. It brings the CNL models trained on the CUB-200, Stanford-Dogs and Stanford-Cars.

Citation

If you think this code is useful in your research or wish to refer to the baseline results published in our paper, please use the following BibTeX entry.

@article{Associating2020Zihan,
    author={Zihan Ye and Fuyuan Hu and Yin Liu and Zhenping Xia and Fan Lyu and Pengqing Liu},
    title={Associating Multi-Scale Receptive Fields for Fine-grained Recognition},
    journal={ICIP},
    year={2020}
}

Requirements

• PyTorch >= 0.4.1 or 1.0 from a nightly release
• Python >= 3.5
• torchvision >= 0.2.1
• termcolor >= 1.1.0

Environment

The code is developed and tested under 1 RTX 8000 GPU cards on CentOS with installed CUDA-9.2/8.0 and cuDNN-7.1.

Baselines and Main Results on CUB-200 Dataset

Model	Best Top-1 (%)	Top-5 (%)
R-50	84.05	96.00
R-50 w/ 5NL	85.10	96.18
R-50 w/ 5CNL	85.64	96.84
R-101	85.05	96.70
R-101 w/ 5NL	85.53	96.65
R-101 w/ 5CNL	86.73	96.75

Notes:

• The input size is 448.
• Prolonging the WARMUP_ITERS appropriately would produce the better results for CNLNet models.

Getting Start

Prepare Dataset

• Download pytorch imagenet pretrained models from pytorch model zoo. The optional download links can be found in torchvision. Put them in the pretrained folder.

• Download the training and validation lists for CUB-200 dataset from Baidu Pan(Password: b8r6).

• Download the training and validation lists for Stanford_Car dataset from Baidu Pan(Password: lrtz).

• Download the training and validation lists for Stanford_Dog dataset from Baidu Pan(Password: ih96). Put them in the data folder and make them look like:

  ${THIS REPO ROOT}
   `-- pretrained
       |-- resnet50-19c8e357.pth
       |-- resnet101-5d3b4d8f.pth
   `-- data
       `-- cub
           `-- images
           |   |-- 001.Black_footed_Albatross
           |   |-- 002.Laysan_Albatross
           |   |-- ...
           |   |-- 200.Common_Yellowthroat
           |-- cub_train.list
           |-- cub_val.list
           |-- images.txt
           |-- image_class_labels.txt
           |-- README
       `-- Stanford_Car
           `-- images
           |-- cars_train.list
           |-- cars_test.list
           |-- ...
       `-- Stanford_Dog
           `-- Images
           |-- dogs_train.list
           |-- dogs_test.list
           |-- ...
  

Perform Validating

$ python train_val.py --arch '50' --dataset 'cub' --checkpoints ${FOLDER_DIR} --valid

Perform Training ResNet with Cross-layer Non-Local module

$ python train_val.py --arch '50' --dataset 'cub' --warmup

License

This code is released under the MIT License. See LICENSE for additional details.

cnl-icip2020's Issues

Question about the result of ResNet-50 baseline?

Hello, Could you please share your pretrained model in these three fine grained datasets?

Hello, I have read your paper, could you please share your pretrained model in these three fine grained datasets? Thank you very much!

Why uses 'SCALE' before softmax?

Looking forward to hearing from you!