This repo contains the official PyTorch implementation of our paper CD-FSOD: A Benchmark for Cross-domain Few-shot Object Detection.

• MS COCO

• ArTaxOr

• UODD

• DIOR

Under the proposed benchmarks, we evaluate existing FSOD methods, including meta-learning methods and fine-tuning methods. We use their official implementation.

• A-RPN [5]
• H-GCN [6]
• Meta-RCNN [7]

• TFA [8]
• FSCE [9]
• DeFRCN [10]

• python >= 3.8
• detectron2 == 0.6
• PyTorch >= 1.10 & torchvision that matches the PyTorch version.
• CUDA==11.3
• GCC >= 5.4

• Clone Code

git clone https://github.com/Paper-ID-1349/CD-FSOD.git
cd CD-FSOD

• Install PyTorch 1.10 with CUDA 11.3

conda install pytorch==1.9.0 torchvision==0.10.0 torchaudio==0.9.0 cudatoolkit=11.3 -c pytorch -c conda-forge

• Install Detectron2

python -m pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/index.html

• Install other requirements.

python -m pip install -r requirements.txt

• Data Preparation

  • Data splits. Download the preprocessed datasets and splits from here(Goodle Drive, or Baidu Wangpan(password: uer8))
  • Unzip the downloaded data-source to datasets and put it into your project directory:

  ...
  datasets
    | -- coco (train2017/*.jpg, val2017/*.jpg, annotations/*.json)
    | -- ArTaxOr (train/*.jpg, test/*.jpg, annotations/*.json)
    | -- UOOD (train/*.jpg, test/*.jpg, annotations/*.json)
    | -- DIOR (train/*.jpg, test/*.jpg, annotations/*.json)
  net
  train_net.py
  ...
  

• Weights Preparation

  • We use the MS COCO pretrain weights to initialize our model. Download the pretrain weights from here (R50-FPN with 3x)

    bash run.sh [dataset]

For example:

    bash run.sh DIOR

This repo is developed based on Detectron2. Please check them for more details and features.

[1] Lin, Tsung-Yi, et al. "Microsoft coco: Common objects in context." ECCV 2014.

[2] https://www.kaggle.com/datasets/mistag/arthropod-taxonomy-orders-object-detection-dataset

[3] Jiang, Lihao, et al. "Underwater species detection using channel sharpening attention." ACM MM 2021.

[4] Li, Ke, et al. "Object detection in optical remote sensing images: A survey and a new benchmark." ISPRS J. Photogramm. Remote Sens. 2020).

[5] Fan, Qi, et al. "Few-shot object detection with attention-RPN and multi-relation detector." CVPR 2020.

[6] Han, Guangxing, et al. "Query adaptive few-shot object detection with heterogeneous graph convolutional networks." ICCV 2021.

[7] Han, Guangxing, et al. "Meta faster r-cnn: Towards accurate few-shot object detection with attentive feature alignment." AAAI 2022.

[8] Wang, Xin, et al. "Frustratingly Simple Few-Shot Object Detection." ICML 2020.

[9] Sun, Bo, et al. "Fsce: Few-shot object detection via contrastive proposal encoding." CVPR 2021.

[10] Qiao, Limeng, et al. "Defrcn: Decoupled faster r-cnn for few-shot object detection." ICCV 2021.