Dark Enhanced Visual Place Recognition

This repository implements a dark-enhanced network for robust visual place recognition in low-light conditions. The project is based on the research paper which introduces ResEM (Residual Enhancement Module) and DSPFormer (Dual-Level Sampling Pyramid Transformer) to enhance image quality and extract discriminative features in challenging environments.

Datasets will be released later.

Table of Contents

• Introduction
• Installation
• Usage
• Training
• Evaluation

Introduction

Visual Place Recognition (VPR) is crucial for mobile robots and autonomous systems, particularly in low-light environments where standard methods struggle. This repository implements the following:

• ResEM (Residual Enhancement Module): A lightweight GAN-based module that enhances image quality under low-light conditions.
• DSPFormer (Dual-Level Sampling Pyramid Transformer): A transformer-based network that extracts robust features for place recognition.

The network is trained using a combination of Triplet Loss and Adversarial Loss to improve performance under challenging conditions.

Installation

To set up the project, follow these steps:

1. Clone the repository:

   git clone https://github.com/CV4RA/Dark-enhanced-VPR-Net.git
   cd Dark-enhanced-VPR-Net/

2. Set up a virtual environment:

   python -m venv vpr_env
   source vpr_env/bin/activate 

3. Install the dependencies:

   pip install -r requirements.txt

Usage

Project Structure

• models/: Contains the implementation of ResEM, DSPFormer, and the loss functions.
• data/: Data loading scripts for training and evaluation.
• train/: Training scripts.
• eval/: Evaluation scripts.
• requirements.txt: Python dependencies.

Training

To train the model, run the following command:

python train_dark_vpr.py

Evaluation

To evaluate the model, run the following command:

python eval_dark_vpr.py