https://github.com/SHI-Labs/Self-Similarity-Grouping

Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification (ICCV 2019, Oral)
https://github.com/SHI-Labs/Self-Similarity-Grouping

computer-vision deep-learning domain-adaptation person-reidentification

Last synced: about 2 months ago
JSON representation

Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification (ICCV 2019, Oral)

• Host: GitHub
• URL: https://github.com/SHI-Labs/Self-Similarity-Grouping
• Owner: SHI-Labs
• Created: 2019-08-03T23:54:00.000Z (over 5 years ago)
• Default Branch: master
• Last Pushed: 2020-05-09T04:12:08.000Z (over 4 years ago)
• Last Synced: 2024-08-10T07:01:17.315Z (4 months ago)
• Topics: computer-vision, deep-learning, domain-adaptation, person-reidentification
• Language: Python
• Homepage:
• Size: 1.89 MB
• Stars: 187
• Watchers: 8
• Forks: 43
• Open Issues: 25
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

• awesome-few-shot-meta-learning - code - official (PyTorch)

README

        
# Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identificatio(SSG)

Implementation of the paper [Self-similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-identification](https://arxiv.org/abs/1811.10144), ICCV 2019 (Oral)

The SSG approach proposed in the paper is simple yet effective and achieves the state-of-arts on three re-ID datasets: Market1501, DukdMTMC and MSMT17.

![Illustration of the Self-similarity Grouping.](./figs/framework.png)

## Running the experiments

### Step 1: Train on source dataset

Run `source_train.py` via

```shell

python source_train.py \

    --dataset \

    --resume \

    --data_dir \

    --logs_dir 

```

To replicate the results in the paper, you can download pre-trained models on Market1501, DukeMTMC and MSMT17 from [GoogleDrive](https://drive.google.com/file/d/1Z94qbsjuAQ9sLeEzURPstQxa3gluZIPJ/view?usp=sharing). There maybe some bugs in source_train.py, please refer to [DomainAdaptiveReID](https://github.com/LcDog/DomainAdaptiveReID) to obtained the pretrained model or just use the pretrained model provided by us.

And you can find all models after adaptation from [GoogleDrive](https://drive.google.com/file/d/1BUp1fbjKTZGjL8WGCx3yUp2GXD10EdxC/view?usp=sharing). Our models can be trained with __PyTorch 0.4.1__ or __PyTorch 1.0__.

### Step 2: Run Self-similarity Grouping

```shell

python selftraining.py \

    --src_dataset \

    --tgt_dataset \

    --resume \

    --iteration \

    --data_dir \

    --logs_dir \

    --gpu-devices \

    --num-split 

```

Or just command

```shell

./run.sh

```

### Step 3: Run Clustering-guided Semi-Supervised Training

```shell

python semitraining.py \

    --src_dataset \

    --tgt_dataset \

    --resume \

    --iteration \

    --data_dir \

    --logs_dir \

    --gpu-devices \

    --num-split \

    --sample 

```

## Results

### Step 1: After training on source dataset

| Source Dataset | Rank-1 | mAP |

| :--- | :---: | :---: |

| DukeMTMC | 82.6 | 70.5 |

| Market1501 | 92.5 | 80.8 |

| MSMT17 | 73.6 | 48.6 |

### Step 2: After adaptation

    

        SRC --> TGT

        Before Adaptation

        Adaptation by SSG

        Adaptation by SSG++

    

    

        Rank-1

        mAP

        Rank-1

        mAP

        Rank-1

        mAP

    

    Market1501 --> DukeMTMC30.516.173.053.476.060.3

    DukeMTMC --> Market150154.626.680.058.386.268.7

    Market1501 --> MSMT17 8.62.731.613.237.616.6

    DukeMTMC --> MSMT17 12.383.8232.213.341.618.3

## Issues

* The pre-trained model is trained with Pytorch 0.4.1, there may be some  error when loading it by Pytorch with higher version. This [link](https://github.com/CSAILVision/places365/issues/25#issuecomment-333871990) should be helpful

* The source_training.py codes may have some bugs, I suggest you directly using our pretrained baseline model. And I will fix the bugs soon.

* To reproduce results listed in paper, I recommend to use two GPUs with batch size of 32. And in general, the experimental results may have be a little different from the results listed in paper (+/-1%).

## Acknowledgement

Our code is based on [open-reid](https://github.com/Cysu/open-reid) and [DomainAdaptiveReID](https://github.com/LcDog/DomainAdaptiveReID).

## Citation

If you find the code helpful in your resarch or work, please cite the following paper.

```

@InProceedings{Fu_2019_ICCV,

author = {Fu, Yang and Wei, Yunchao and Wang, Guanshuo and Zhou, Yuqian and Shi, Honghui and Huang, Thomas S.},

title = {Self-Similarity Grouping: A Simple Unsupervised Cross Domain Adaptation Approach for Person Re-Identification},

booktitle = {The IEEE International Conference on Computer Vision (ICCV)},

month = {October},

year = {2019}

}

```