FewX

FewX is an open source toolbox on top of Detectron2 for data-limited instance-level recognition tasks, e.g., few-shot object detection, few-shot instance segmentation, partially supervised instance segmentation and so on.

All data-limited instance-level recognition works from Qi Fan (HKUST, [email protected]) are open-sourced here.

To date, FewX implements the following algorithms:

• FSOD: few-shot object detection with FSOD dataset.
• CPMask: partially supervised/fully supervised/few-shot instance segmentation. (20220725 working on it)
• FSVOD: few-shot video object detection with FSVOD-500 dataset and FSYTV-40 dataset. (20220725 working on it)

Highlights

• State-of-the-art performance.
  • FSOD is the best few-shot object detection model. (This model can be directly applied to novel classes without finetuning. And finetuning can bring better performance.)
  • CPMask is the best partially supervised/few-shot instance segmentation model.
• Easy to use. You only need to run 3 code lines to conduct the entire experiment.
  • Install Pre-Built Detectron2 in one code line.
  • Prepare dataset in one code line. (You need to first download the dataset and change the data path in the script.)
  • Training and evaluation in one code line.

Updates

• FewX has been released. (09/08/2020)

Results on MS COCO

Few Shot Object Detection

Method	Training Dataset	Evaluation way&shot	box AP	download
FSOD (paper)	COCO (non-voc)	full-way 10-shot	11.1	-
FSOD (this implementation)	COCO (non-voc)	full-way 10-shot	12.0	model | metrics

The results are reported on the COCO voc subset with ResNet-50 backbone.

The model only trained on base classes is base model .

You can reference the original FSOD implementation on the Few-Shot-Object-Detection-Dataset.

Step 1: Installation

You only need to install detectron2. We recommend the Pre-Built Detectron2 (Linux only) version with pytorch 1.7. I use the Pre-Built Detectron2 with CUDA 10.1 and pytorch 1.7 and you can run this code to install it.

python -m pip install detectron2 -f \
  https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/torch1.7/index.html

Step 2: Prepare dataset

• Prepare for coco dataset following this instruction.

• cd datasets, change the DATA_ROOT in the generate_support_data.sh to your data path and run sh generate_support_data.sh.

cd FewX/datasets
sh generate_support_data.sh

Step 3: Training and Evaluation

Run sh all.sh in the root dir. (This script uses 4 GPUs. You can change the GPU number. If you use 2 GPUs with unchanged batch size (8), please halve the learning rate.)

cd FewX
sh all.sh

TODO

• Add FSVOD and CPMask codes to this repo.
• Add other dataset results to FSOD.
• Add CPMask code with partially supervised instance segmentation, fully supervised instance segmentation and few-shot instance segmentation.

Citing FewX

If you use this toolbox in your research or wish to refer to the baseline results, please use the following BibTeX entries.

@inproceedings{fan2021fsvod,
  title={Few-Shot Video Object Detection},
  author={Fan, Qi and Tang, Chi-Keung and Tai, Yu-Wing},
  booktitle={ECCV},
  year={2022}
}
@inproceedings{fan2020cpmask,
  title={Commonality-Parsing Network across Shape and Appearance for Partially Supervised Instance Segmentation},
  author={Fan, Qi and Ke, Lei and Pei, Wenjie and Tang, Chi-Keung and Tai, Yu-Wing},
  booktitle={ECCV},
  year={2020}
}
@inproceedings{fan2020fsod,
  title={Few-Shot Object Detection with Attention-RPN and Multi-Relation Detector},
  author={Fan, Qi and Zhuo, Wei and Tang, Chi-Keung and Tai, Yu-Wing},
  booktitle={CVPR},
  year={2020}
}

Special Thanks

Detectron2, AdelaiDet, centermask2