This is the official implementation of PF-RPN, a state-of-the-art model for Cross-Domain Open-Set Region Proposal generation, accepted at CVPR 2026.

Paper | GitHub Repository

🧠 Model Details

PF-RPN (Prompt-Free Universal Region Proposal Network) identifies potential objects without relying on external prompts (like class names, exemplar images, or textual descriptions). Instead of specific category names, all categories are unified into a single learnable token (object).

Model Architecture Innovations

To improve proposal quality without explicit class prompts, PF-RPN introduces three key designs:

Sparse Image-Aware Adapter (SIA): Performs initial localization of potential objects using a learnable query embedding dynamically updated with visual features.
Cascade Self-Prompt (CSP): Identifies remaining objects by leveraging self-prompted learnable embeddings, autonomously aggregating informative visual features in a cascading manner.
Centerness-Guided Query Selection (CG-QS): Facilitates the selection of high-quality query embeddings using a centerness scoring network.

Model Sources

Repository: PF-RPN GitHub Repository
Base Framework: MMDetection 3.3.0
Backbone: Swin-Base (swinb)

🎯 Intended Use

Primary Use Case: Generating high-quality, class-agnostic region proposals ("objects") across diverse, unseen domains without requiring domain-specific text prompts or retraining.
Applications: Underwater object detection, industrial defect detection, and remote sensing image object detection.

🗂️ Training Data

The provided checkpoint (pf_rpn_swinb_5p_coco_imagenet.pth) was trained on a combined dataset of COCO 2017 (5% subset) and ImageNet-1k (5% subset).

To simulate the open-set proposal generation task, all ground-truth categories are merged into a single class (object).

📊 Performance

PF-RPN achieves state-of-the-art Average Recall (AR) under prompt-free evaluation across multiple benchmarks.

Cross-Domain Few-Shot Object Detection (CD-FSOD)

Method	Prompt Free	AR100	AR300	AR900
GDINO‡	✓	54.7	57.8	61.6
GenerateU	✓	47.7	54.1	55.7
PF-RPN (Ours)	✓	60.7	65.3	68.2

Object Detection in the Wild (ODinW13)

Method	Prompt Free	AR100	AR300	AR900
GDINO‡	✓	69.1	70.9	72.4
GenerateU	✓	67.3	71.5	72.2
PF-RPN (Ours)	✓	76.5	78.6	79.8

(‡ indicates models where original class names were replaced with object to simulate a prompt-free setting).

⚙️ How to Use

Installation

The codebase is built on MMDetection. Please follow the installation instructions in the official repository.

Quick Start: Evaluation

Download the Weights

bash

mkdir -p checkpoints

# Download GroundingDINO base weights
wget -O checkpoints/groundingdino_swinb_cogcoor_mmdet-55949c9c.pth \
  https://download.openmmlab.com/mmdetection/v3.0/grounding_dino/groundingdino_swinb_cogcoor_mmdet-55949c9c.pth

# Download PF-RPN weights
wget -O checkpoints/pf_rpn_swinb_5p_coco_imagenet.pth \
  https://huggingface.co/tangqh/PF-RPN/resolve/main/pf_rpn_swinb_5p_coco_imagenet.pth

Run Testing

bash

python tools/test.py \
  configs/pf-rpn/pf-rpn_coco-imagenet.py \
  checkpoints/pf_rpn_swinb_5p_coco_imagenet.pth

📚 Citation

If you use PF-RPN in your research, please cite:

bibtex

@inproceedings{tang2026pf,
  title={Prompt-Free Universal Region Proposal Network},
  author={Tang, Qihong and Liu, Changhan and Zhang, Shaofeng and Li, Wenbin and Fan, Qi and Gao, Yang},
  booktitle={CVPR},
  year={2026}
}