SCSM

KO | EN

gitlite — search

#javascript #python #react #typescript #go #rust #macos #windows #awesome #cli #vue #pytorch #security #mcp

(★ 10)

No description available.

SCSM Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

## Few-Shot Object Detection via Spatial-Channel State Space Model

## Quick Start

**1. Check Requirements**
* Linux with Python >= 3.10
* [PyTorch](https://pytorch.org/get-started/locally/) >= 2.1.1 & [torchvision](https://github.com/pytorch/vision/) that matches the PyTorch version.
* CUDA 11.7 11.8

**2. Build SCSM**

* Install PyTorch 2.1.1 with CUDA 11.8 
  ```shell
  conda install pytorch==2.1.1 torchvision==0.16.1 torchaudio==2.1.1 pytorch-cuda=11.8 -c pytorch -c nvidia
  ```
* Install Detectron2==0.6
 
  - If you use other version of PyTorch/CUDA, check the latest version of Detectron2 in this page: [Detectron2](https://github.com/facebookresearch/detectron2/releases). 
 
* Install other requirements. 
  ```angular2html
  pip install -r requirements.txt
  ```

**3. Prepare Data**
* Data Preparation
 
  - Unzip the downloaded data-source to `datasets` and put it into your project directory:
    ```angular2html
      ...
      datasets
        | -- coco (trainval2014/*.jpg, val2014/*.jpg, annotations/*.json)
        | -- cocosplit
        | -- VOC2007
        | -- VOC2012
        | -- vocsplit
      scsm
      tools
      ...
    ```

**4. Training and Evaluation**

* Fist run voc base.
  ```angular2html
  bash run_voc_base.sh    
  ```

* pre-trained base model prepare.
  ```angular2html
   python tools/model_surgery.py --dataset voc --method randinit                         \
     --src-path ${SAVE_DIR}/voc_base/model_final.pth                      \
     --save-dir ${SAVE_DIR}/voc_base
  ```

* Then run voc novel.
  ```angular2html
  bash run_voc_novel.sh    # ATTACK represents AFT.
  ```

* Fist run coco base.
  ```angular2html
  bash run_coco_base.sh
  ```

* pre-trained base model prepare.
  ```angular2html
   python tools/model_surgery.py --dataset voc --method remove                         \
     --src-path ${SAVE_DIR}/coco_base/model_final.pth                      \
     --save-dir ${SAVE_DIR}/coco_base
  ```

* Then run coco novel.
  ```angular2html
  bash run_coco_novel.sh  # ATTACK represents AFT.
  ```

## Citing

If you use this work in your research or wish to refer to the baseline results published here, please use the following BibTeX entries:
```

@ARTICLE{ECEA,
  author={Xin, Zhimeng and Wu, Tianxu and Chen, Shiming and Zou, Yixiong and Shao, Ling and You, Xinge},
  journal={IEEE Transactions on Image Processing}, 
  title={ECEA: Extensible Co-Existing Attention for Few-Shot Object Detection}, 
  year={2024},
  volume={33},
  pages={5564-5576},
  doi={10.1109/TIP.2024.3411771}}

@ARTICLE{SCSM,
  author={Xin, Zhimeng and Wu, Tianxu and Zou, Yixiong and Chen, Shiming and Fu, Dingjie and You, Xinge},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Few-Shot Object Detection via Spatial-Channel State Space Model}, 
  year={2026},
  volume={36},
  number={5},
  pages={6273-6283},
  doi={10.1109/TCSVT.2025.3642750}}

@ARTICLE{AFT,
  author={Wu, Tianxu and Xin, Zhimeng and Chen, Shiming and Zou, Yixiong and You, Xinge},
  journal={IEEE Transactions on Circuits and Systems for Video Technology}, 
  title={Adversarial Feature Training for Few-Shot Object Detection}, 
  year={2025},
  volume={35},
  number={9},
  pages={9324-9336},  
  doi={10.1109/TCSVT.2025.3552138}}
```

SCSM

SCSM Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

SCSM

SCSM Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README