Awesome

DR-Tune

https://arxiv.org/abs/2308.12058

This repository is an official PyTorch implementation of DR-Tune: Improving Fine-tuning of Pretrained Visual Models by Distribution Regularization with Semantic Calibration (ICCV2023).

Usage

Environments

python 3.9.7
pytorch 1.13.1
torchvision 0.14.1
GPU NVIDIA GeForce RTX 2080 Ti

Dataset preparation

The datasets used in Table 1 can be downloaded via their official link.

The datasets used in Table 2 can be downloaded from here and see "vtab-1k".

Pretrained model preparation

The pretrained model checkpoints used in the paper can be found in the table below.

Please put the checkpoint in ./pretrained_models.

Backbone architecture	Pretraining strategy	Url
ViT-B	Classification	Checkpoint
ViT-B	MAE	Checkpoint
ViT-L	MAE	Checkpoint
ResNet-50	MoCo-v1	Checkpoint
ResNet-50	MoCo-v2	Checkpoint
ResNet-50	PCL	Checkpoint
ResNet-50	HCSC	Checkpoint
ResNet-50	SwAV	Checkpoint
ResNet-50/101/152	InfoMin	Checkpoint
ResNeXt-101/152	InfoMin	Checkpoint

Training

Fine-tuning a ResNet-50 pretrained by MoCo-v2 on CIFAR10.

CIFAR10 will be automatically downloaded to ./data.

bash train.sh 1 --cfg ./configs/cifar10_k2048_lr001.yaml

Citation

If you find our work helpful in your research, please cite it as:

@inproceedings{zhou2023dr,
  title={DR-Tune: Improving Fine-tuning of Pretrained Visual Models by Distribution Regularization with Semantic Calibration},
  author={Zhou, Nan and Chen, Jiaxin and Huang, Di},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={1547--1556},
  year={2023}
}