AGPF-FSFG

Official codes for Learning Attention-Guided Pyramidal Features for Few-shot Fine-grained Recognition (Pattern Recognition, 2022)

By Hao Tang, Chengcheng Yuan, Zechao Li, and Jinhui Tang

Extension of Conference Paper (IJCAI 2021 LTDL Workshop Best Paper Award)

Enviroment

• Python3
• Pytorch >= 1.6.0
• CUDA = 10.2
• json

Datasets

CUB

• Change directory to ./filelists/CUB
• run source ./download_CUB.sh

FGVC

• Change directory to ./filelists/Aircrafts
• change variable data_dir in Aircrafts_prepare_csv.py
• run python Aircrafts_prepare_csv.py

StanfordCars

• Change directory to ./filelists/StanfordCars
• change variable data_dir in StanforCar_prepare_csv.py
• run python StanforCar_prepare_csv.py

StanfordDogs

• Change directory to ./filelists/StanfordDogs
• change variable data_dir in StanfordDog_prepare_csv.py
• run python StanfordDog_prepare_csv.py

Self-defined setting

• Require three data split json file: base.json, val.json, novel.json for each dataset
• The format should follow
  {"label_names": ["class0","class1",...], "image_names": ["filepath1","filepath2",...],"image_labels": [l1,l2,l3,...]}
  See test.json for reference
• Put these file in the same folder and change data_dir ['DATASETNAME'] in configs.py to the folder path

Train

Run
python ./train.py --train_n_way [TRAIN_N_WAY] --test_n_way [TEST_N_WAY] --n_shot [K_SHOT] --stop_epoch [EPOCHS] --dataset [DATASETNAME] --model [BACKBONENAME] --method [METHODNAME] --num_classes [NUM_CLASSES] --train_aug --apcnn [--OPTIONARG]

For example,
python ./train.py --train_n_way 5 --test_n_way 5 --n_shot 1 --stop_epoch 120 --dataset CUB --model Conv4 --method protonet --num_classes 200 --train_aug --apcnn

Commands below follow this example, and please refer to io_utils.py for additional options.

Save features

Save the extracted feature before the classifaction layer to increase test speed.
python ./save_features.py --train_n_way 5 --test_n_way 5 --n_shot 1 --dataset CUB --model Conv4 --method protonet --train_aug --apcnn --num_classes 200

Test

python ./test.py --train_n_way 5 --test_n_way 5 --n_shot 1 --dataset CUB --model Conv4 --method protonet --train_aug --apcnn --num_classes 200

Results

• The test results will be recorded in ./record/results.txt

Other related papers

   @inproceedings{TianTD21,
    author    = {Sheng Tian and Hao Tang and Longquan Dai},
    title     = {Coupled Patch Similarity Network FOR One-Shot Fine-Grained Image Recognition},
    booktitle = {ICIP},
    pages     = {2478--2482},
    year      = {2021}，
    publisher = {{IEEE}}
  }

   @article{zha2023boosting,
     title={Boosting Few-shot Fine-grained Recognition with Background Suppression and Foreground Alignment},
     author={Zha, Zican and Tang, Hao and Sun, Yunlian and Tang, Jinhui},
     journal={IEEE Transactions on Circuits and Systems for Video Technology},
     year={2023},
     publisher={IEEE},
     doi={10.1109/TCSVT.2023.3236636}
  }

Citation

If this work is useful in your research, please cite

   @article{TangYLT22,
        author    = {Hao Tang and Chengcheng Yuan and Zechao Li and Jinhui Tang},
        title     = {Learning attention-guided pyramidal features for few-shot fine-grained recognition},
        journal   = {Pattern Recognit.},
        volume    = {130},
        pages     = {108792},
        year      = {2022}
   }

References

This implementation builds upon several open-source codes. Specifically, we have modified and integrated the following codes into this repository:

• CloserLookFewShot
• AP-CNN