PCL-CLIP

KO | EN

gitlite — search

#javascript #react #android #deep-learning #linux #windows #awesome #kubernetes #awesome-list #pytorch #ios #mcp #openai #open-source

(★ 37)

Code for "Prototypical Contrastive Learning-based CLIP Fine-tuning for Object Re-identification".

PCL-CLIP 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# Prototypical Contrastive Learning-based CLIP Fine-tuning for Object Re-identification

[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/prototypical-contrastive-learning-based-clip/unsupervised-person-re-identification-on-12)](https://paperswithcode.com/sota/unsupervised-person-re-identification-on-12?p=prototypical-contrastive-learning-based-clip) \
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/prototypical-contrastive-learning-based-clip/unsupervised-vehicle-re-identification-on)](https://paperswithcode.com/sota/unsupervised-vehicle-re-identification-on?p=prototypical-contrastive-learning-based-clip) \
[![PWC](https://img.shields.io/endpoint.svg?url=https://paperswithcode.com/badge/prototypical-contrastive-learning-based-clip/unsupervised-person-re-identification-on-4)](https://paperswithcode.com/sota/unsupervised-person-re-identification-on-4?p=prototypical-contrastive-learning-based-clip)

[[paper]](https://arxiv.org/pdf/2310.17218.pdf)

In this work, we propose a simple yet effective approach to adapt CLIP for supervised Re-ID, which directly fine-tunes the image encoder of CLIP using a Prototypical Contrastive Learning loss. Experimental results demonstrate the simplicity and competitiveness of our method compared to recent prompt-learning-based [CLIP-ReID](). Futhermore, our investigation indicates the essential consistency between the CLIP-ReID and our method.

## Upload History

* 2023/11/23: Full model (ID loss + CC loss) is ready.

## Pipeline
![pipeline](https://raw.githubusercontent.com/RikoLi/PCL-CLIP/main/assets/pcl-clip-pipeline.jpg)

## Installation

Install `conda` before installing any requirements.

```bash
conda create -n pclclip python=3.9
conda activate pclclip
pip install -r requirements.txt
```

## Datasets

Make a new folder named `data` under the root directory. Download the datasets and unzip them into `data` folder.
* [Market1501](https://drive.google.com/file/d/0B8-rUzbwVRk0c054eEozWG9COHM/view)
* [MSMT17](https://arxiv.org/abs/1711.08565)
* [VeRi776](https://github.com/JDAI-CV/VeRidataset)

## Training

For example, training the full model on Market1501 with GPU 0 and saving the log file and checkpoints to `logs/market-pclclip`:

```
CUDA_VISIBLE_DEVICES=0 python train_pcl.py --config_file config/pcl-vit.yml DATASETS.NAMES "('market1501')" OUTPUT_DIR logs/market-pclclip
```

Configs can be modified from `config/*.yaml` files or from command line. If you want to add more config terms, update `config/defaults.py`. For other models using different losses, please modify the code according to the paper.

## Acknowledgement

The code is implemented based on following works. We sincerely appreciate their great efforts on Re-ID research!

1. [TransReID](https://github.com/damo-cv/TransReID)
2. [CLIP-ReID](https://github.com/Syliz517/CLIP-ReID)
3. [ClusterContrast](https://github.com/alibaba/cluster-contrast-reid)
4. [CAP](https://github.com/Terminator8758/CAP-master)
5. [O2CAP](https://github.com/Terminator8758/O2CAP)

## Citation
```bib
@article{li2023prototypical,
  title={Prototypical Contrastive Learning-based CLIP Fine-tuning for Object Re-identification},
  author={Li, Jiachen and Gong, Xiaojin},
  journal={arXiv preprint arXiv:2310.17218},
  year={2023}
}
```

PCL-CLIP

PCL-CLIP 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

PCL-CLIP

PCL-CLIP 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README