RALF

Official implementation of CVPR 2024 paper "Retrieval-Augmented Open-Vocabulary Object Detection".

Introduction

RALF is structured into multiple branches.

• prerequisite branch: The code for prerequisites necessary for running RALF.
• RAF branch: The code for training RAF.

The other branches are the integration of existing OVD model and RALF.

• OADP branch: The baseline is OADP.
• Centric branch: The baseline is Object-Centric-OVD.
• DetPro branch: The baseline is DetPro.

Results

COCO

Model	$\text{AP}^\text{N}_\text{50}$
RALF + OADP	33.4
RALF + Object-Centric-OVD	41.3

LVIS

Model	$\text{AP}_\text{r}$
RALF + OADP	21.9
RALF + DetPro	21.1
RALF + Object-Centric-OVD	18.5

Citation

@inproceedings{kim2024retrieval,
  title={Retrieval-Augmented Open-Vocabulary Object Detection},
  author={Kim, Jooyeon and Cho, Eulrang and Kim, Sehyung and Kim, Hyunwoo J},
  booktitle={CVPR},
  year={2024}
}

References

This code is built on CLIP, V3Det, GPT-3, OADP, Object-Centric-OVD and DetPro.