Generalized Few-Shot Object Detection without Forgetting

This project provides an implementation for "Generalized Few-Shot Object Detection without Forgetting" (CVPR2021) on PyTorch.

Experiments in the paper were conducted on the internal framework, thus we reimplement them on cvpods and report details as below.

Requirements

• cvpods

Get Started

• install cvpods locally (requires cuda to compile)

python3 -m pip install 'git+https://github.com/Megvii-BaseDetection/cvpods.git'
# (add --user if you don't have permission)

# Or, to install it from a local clone:
git clone https://github.com/Megvii-BaseDetection/cvpods.git
python3 -m pip install -e cvpods

# Or,
pip install -r requirements.txt
python3 setup.py build develop

• prepare datasets

cd /path/to/cvpods
cd datasets
ln -s /path/to/your/coco/dataset coco
ln -s /path/to/your/voc/dataset voc

• prepare datasets of few-shot detection benchmarks as "FSDET_DATASET.md"

• prepare pretrained weights

# download the pretrained faster r-cnn of TFA
# at http://dl.yf.io/fs-det/models/
# or train the weights yourself

# modify the weights' path in register_pretrained_weights.sh then
bash register_pretrained_weights.sh

• Train & Test (default w/ 8 GPUs)

cd /path/to/experiment
bash run.sh

Results on COCO few-shot detection benchmarks

experiment	AP	bAP	nAP
5shot	31.5	39.2	8.3
10shot	32.1	39.2	10.5
30shot	32.9	39.3	13.8

Results on VOC few-shot detection benchmarks

experiment	AP50	bAP50	nAP50
split1-1shot	71.3	80.9	42.4
split1-2shot	72.3	81.1	45.8
split1-3shot	72.1	80.8	45.9
split1-5shot	74.0	80.8	53.7
split1-10shot	74.6	80.8	56.1
split2-1shot	66.8	81.8	21.7
split2-2shot	68.4	81.9	27.8
split2-3shot	70.2	81.9	35.2
split2-5shot	70.7	81.9	37.0
split2-10shot	71.5	81.9	40.3
split3-1shot	69.0	81.9	30.2
split3-2shot	70.9	82.0	37.6
split3-3shot	72.3	82.1	43.0
split3-5shot	73.9	82.0	49.7
split3-10shot	74.1	82.1	50.1

Acknowledgement

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

License

This repo is released under the Apache 2.0 license. Please see the LICENSE file for more information.

Citing

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

@article{fan2021generalized,
  title     =  {Generalized Few-Shot Object Detection without Forgetting},
  author    =  {Fan, Zhibo and Ma, Yuchen and Li, Zeming and Sun, Jian},
  booktitle =  {IEEE Conference on Computer Vision and Pattern Recognition},
  year      =  {2021}
}

Contributing to the project

Any pull requests or issues about the implementation are welcome. If you have any issue about the library (e.g. installation, environments), please refer to cvpods.