Awesome
MDCS: More Diverse Experts with Consistency Self-distillation for Long-tailed Recognition [Official, ICCV 2023, Paper] š„
Qihao Zhao<sup>1,2</sup>, Chen Jiang<sup>1</sup>, Wei Hu<sup>1</sup>, Fan Zhang<sup>1</sup>, Jun Liu<sup>2</sup>
1 Beijing University of Chemical Technology
2 Singapore University of Technology and Design
0.Citation
If you find our work inspiring or use our codebase in your research, please consider giving a star ā and a citation.
@InProceedings{Zhao_2023_ICCV,
author = {Zhao, Qihao and Jiang, Chen and Hu, Wei and Zhang, Fan and Liu, Jun},
title = {MDCS: More Diverse Experts with Consistency Self-distillation for Long-tailed Recognition},
booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
month = {October},
year = {2023},
pages = {11597-11608}
}
1.training
(1) CIFAR100-LT
Training
- run:
python train.py -c configs/config_cifar100_ir100_mdcs.json
Evaluate
- run:
python test.py -r checkpoint_path
(2) ImageNet-LT
Training
- run such as resnext50 400 epochs:
python train.py -c configs/config_imagenet_lt_resnext50_mdcs_e400.json
Evaluate
- run:
python test.py -r checkpoint_path
(3) Places-LT
Training
- run:
python train_places.py -c configs/config_places_lt_resnet152_mdcs.json
Evaluate
- run:
python test_places.py -r checkpoint_path
(4) iNaturalist 2018
Training
- run :
python train.py -c configs/config_iNaturalist_resnet50_mdcs.json
Evaluate
- run:
python test.py -r checkpoint_path
2. Requirements
- To install requirements:
pip install -r requirements.txt
- Run in linux (may have some problems in windows)
3. Datasets
(1) Four bechmark datasets
- Please download these datasets and put them to the /data file.
- ImageNet-LT and Places-LT can be found at here.
- iNaturalist data should be the 2018 version from here.
- CIFAR-100 will be downloaded automatically with the dataloader.
data
āāā ImageNet_LT
āĀ Ā āāā test
āĀ Ā āāā train
āĀ Ā āāā val
āāā CIFAR100
āĀ Ā āāā cifar-100-python
āāā Place365
āĀ Ā āāā data_256
āĀ Ā āāā test_256
āĀ Ā āāā val_256
āāā iNaturalist
Ā Ā āāā test2018
āāā train_val2018
(2) Txt files
- We provide txt files for test-agnostic long-tailed recognition for ImageNet-LT, Places-LT and iNaturalist 2018. CIFAR-100 will be generated automatically with the code.
- For iNaturalist 2018, please unzip the iNaturalist_train.zip.
data_txt
āāā ImageNet_LT
āĀ Ā āāā ImageNet_LT_test.txt
āĀ Ā āāā ImageNet_LT_train.txt
āĀ Ā āāā ImageNet_LT_val.txt
āāā Places_LT_v2
āĀ Ā āāā Places_LT_test.txt
āĀ Ā āāā Places_LT_train.txt
āĀ Ā āāā Places_LT_val.txt
āāā iNaturalist18
āāā iNaturalist18_train.txt
āāā iNaturalist18_uniform.txt
āāā iNaturalist18_val.txt
4. Pretrained models
- For the training on Places-LT, we follow previous methods and use the pre-trained ResNet-152 model.
- Please download the checkpoint. Unzip and move the checkpoint files to /model/pretrained_model_places/.
5. Acknowledgements
The mutli-expert framework is based on SADE and RIDE. Strong augmentations are based on NCL and PaCo.