This repository contains the source code for our paper "Retentive Compensation and Personality Filtering for Few-Shot Remote Sensing Object Detection" by Jiashan Wu, Chunbo Lang, Gong Cheng, Xingxing Xie, and Junwei Han.
1. Check Requirements
- Linux with Python >= 3.8
- PyTorch >= 1.6 & torchvision that matches the PyTorch version.
- CUDA 10.1
- GCC >= 4.9
2. Build RP-FSOD
- Clone Code
git clone https://github.com/yomik-js/RP-FSOD.git cd RP-FSOD - Create a virtual environment (optional)
conda create -n RP-FSOD python==3.8.0 conda activate RP-FSOD - Install PyTorch 1.7.1 with CUDA 10.1
pip install torch==1.7.1+cu101 torchvision==0.8.2+cu101 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
- Install Detectron2
python3 -m pip install detectron2==0.3 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu101/torch1.7/index.html- If you use other version of PyTorch/CUDA, check the latest version of Detectron2 in this page: Detectron2.
- Sorry for that I don’t have enough time to test on more versions, if you run into problems with other versions, please let me know.
- Install other requirements.
python3 -m pip install -r requirements.txt
3. Prepare Data and Weights
- Data Preparation
- We evaluate our models on DIOR and NWPU VHR-10.v2, put them into
datasetsand put it into your project directory:... datasets | -- DIOR (JPEGImages/*.jpg, ImageSets/, Annotations/*.xml) | -- DIORsplit | -- NWPU | -- NWPUsplit RP-FSOD tools ...
- We evaluate our models on DIOR and NWPU VHR-10.v2, put them into
- Weights Preparation
- We use the imagenet pretrain weights to initialize our model. Download the same models from here: GoogleDrive
4. Training and Evaluation
For ease of training and evaluation over multiple runs, we integrate the whole pipeline of few-shot object detection into one script run_*.sh, including base pre-training and novel-finetuning.
- To reproduce the results on VOC,
EXP_NAMEcan be any string andSPLIT_IDmu 558B st be1 or 2 or 3 or 4.bash run_voc.sh EXP_NAME SPLIT_ID - Please read the details of few-shot object detection pipeline in
run_*.sh, you need changeIMAGENET_PRETRAIN*to your path.
This repo is developed based on DeFRCN. Please check it for more details and features.