Skip to content

ai4ce/LoQI-VPR

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

2 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

LoQI-VPR

VPR for Low Quality Images via Knowledge Distillation

Implementation for the RAL submission "Distillation Improves Visual Place Recognition for Low Quality Images".

Setup

  1. Clone the repository with submodules: $ git clone --recurse-submodules https://github.com/ai4ce/LoQI-VPR.git
  2. Install dependencies: $ conda env create -f environment.yml
  3. Download GSV-Cities dataset from Kaggle and the Pitts250k dataset for validation
  4. Download VPR testing datasets using VPR Datasets Downloader

Running Experiments

trainer.yaml and test_trained_model.yaml from configs contains the configurations for running distillation and testing VPR methods respectively.

Training: src/trainer_gsv-cities.py distills VPR models enabled in configurations using the enabled loss functions.

Testing: src/dataset/testing_data.py precomputes global descriptors for the specified VPR models and datasets selected in configurations. src/calculate_recall.py records recall rates to tensorboard log files and a Google Sheet.

About

For more information, please visit the project website

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages