Introduction

This reposity is codebase for ACM Multimedia 2022 paper "Exploring Effective Knowledge Transfer for Few-shot Object Detection" .

Code is based on MMFewshot. For installing MMFewshot, please refer to this page

Requirements

1. MMFewshot installed: https://github.com/open-mmlab/mmfewshot

2. my requirements ( just for your checking, theoretically mmfewshot installed is enough. )

python == 3.7.11
torch == 1.7.0
torchvision == 0.8.0
cuda == 10.1
mmfewshot == 0.1.0

Data Preparation

PASCAL VOC

For PASCAL VOC data, please refer to GoogleDrive. Put the archieve under data/ and decompress it.

MS COCO

few-shot annotation

MMFewshot using few-shot split file prepared in advance. Download GoogleDrive and also put decompressed folder under data/.

Model Inference

MS COCO

1. download model from link provided and put model under corresponding folder: ./eval_configs/coco/kshot/
2. run the following command

   CUDA_VISIBLE_DEVICES=0 python ./tools/detection/test.py \
   ./eval_configs/coco/kshot/config.py \
   ./eval_configs/coco/kshot/best.pth --eval bbox

Model Training

• [ ] To be done