Beyond Duality: A Hybrid Framework of Leveraging Shared and Private Features for RGB-Event Object Detection
CVPR 2026
Our paper is available on the CVF Open Access:
This repository provides the official implementation of this paper.
- [2026.06] 🎉 Our paper has been accepted by CVPR 2026.
- [2026.06] 🚀 The official implementation of SPFD is released.
SPFD is designed for RGB-Event object detection. It leverages both shared and private modality-specific features to better exploit the complementary information between RGB frames and event streams.
We tested our code with Python 3.8, PyTorch 1.12.0, and CUDA 11.3.
Our code is based on detrex. Please also refer to the official detrex installation guide.
# Create a new virtual environment
conda create -n SPFD python=3.8 -y
conda activate SPFD
# Install PyTorch
pip install torch==1.12.0+cu113 torchvision==0.13.0+cu113 torchaudio==0.12.0 --extra-index-url https://download.pytorch.org/whl/cu113
# Clone this repository
git clone https://github.com/git-KeYw/SPFD.git
cd SPFD
# Initialize the detectron2 submodule
git init
git submodule init
git submodule update
# Install detectron2
python -m pip install 'git+https://github.com/facebookresearch/detectron2.git'
# Install other dependencies
pip install -r requirements.txt
# Build an editable version of detrex
pip install -e .Please configure the dataset root path for PKU-DAVIS-SOD in:
detrex/data/datasets/register_pku.py
Specifically, modify the dataset path around:
Line 454
Then move the dataset config file:
mv ./projects/midetr/configs/data/pku.py ./detrex/config/configs/common/dataYou can train SPFD on the PKU-DAVIS-SOD dataset with:
python tools/train_net.py \
--num-gpus 2 \
--config-file projects/midetr/configs/midetr-resnet/SPFD_pku.pyYou can evaluate the pretrained model with:
python tools/test_net.py \
--num-gpus 2 \
--config-file projects/midetr/configs/midetr-resnet/SPFD_pku.pyIf you find this work useful for your research, please consider citing our paper:
@InProceedings{Wang_2026_CVPR,
author = {Wang, Keyao and Liu, Shuai and Shi, Hengda and Shi, Lukui and Chen, Haiyong},
title = {Beyond Duality: A Hybrid Framework of Leveraging Shared and Private Features for RGB-Event Object Detection},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2026},
pages = {4415-4424}
}This project is built upon the excellent open-source projects:
We sincerely thank the authors for their great work.
This project is released for academic research purposes only.
