PF-BCP

Project Page | Video Overview | Paper Preprint

This repository contains the official code for training and evaluating the methods as described in:

Potential Field as Scene Affordance for Behavior Change-Based Visual Risk Object Identification
Pang-Yuan Pao, Shu-Wei Lu, Ze-Yan Lu and Yi-Ting Chen
National Yang Ming Chiao Tung University

_{Visual Risk Object Identification by different methods.
All detected risk objects are shown with green bounding boxes, while ground truth risks are masked in red.
BS, PF refer to Bird's-Eye-View Segmentation and Potential Field respectively.}

🔥 Update

• October 2024: 📊 ROI demo tool released!
• September 2024: 🎉 Project page and YouTube video announced!

⚙️ Getting Started

📝 System Setup

• Operating System: Linux Ubuntu 18.04
• Python Version: 3.7
• PyTorch Version: 1.8.0
• CUDA Version: 11.6
• GPU: Nvidia RTX 3090
• CPU: Intel Core i7-10700

📥 Dependency Installation

1. Clone the Repository

   git clone https://github.com/HCIS-Lab/PF-BCP

2. Create and activate a new Conda environment:

   conda create -n YOUR_ENV python=3.7
   conda activate YOUR_ENV
   cd PF-BCP

3. Run the following command to install all required packages from requirements.txt:

   pip install -r requirements.txt

📦 Datasets Downloads

• Download RiskBench_Dataset here.

• Download metadata.zip here.

🚀 Usage

Comming Soon

📊 ROI Demo

We perform offline risk object identification evaluations and visualize fine-grained scenario-based analysis using preserved prediction data as input.

To perform the evaluation, please follow the instructions provided here.

📄 BibTeX

If our work contributes to your research, please consider citing it with the following BibTeX entry:

@article{pao2024PFBCP,
    title   = {{Potential Field as Scene Affordance for Behavior Change-Based Visual Risk Object Identification}},
    author  = {Pang-Yuan Pao and Shu-Wei Lu and Ze-Yan Lu and Yi-Ting Chen},
    year    = {2024},
    eprint  = {2409.15846},
    archivePrefix = {arXiv}
}

🙌 Acknowledgment

We acknowledge that the dataset and baselines used in this project are adapted from RiskBench.