Contrastive Learning-based Place Descriptor Representation for Cross-modality Place Recognition

This repo contains the details for "Contrastive Learning-based Place Descriptor Representation for Cross-modality Place Recognition".

🔑 Set up

Our experiment is tested on Ubuntu 20.04 with Python 3.8 with PyTorch 1.13.6.

• build environment

  conda create -n tmnet python=3.8
  conda activate tmnet
  pip install -r requirements.txt
  

📚 DataSet

We conduct the image-to-point-cloud place recognition based on KITTI dataset and KITTI-360 dataset.

• The KITTI DATASET The used data for the experiment can be downloaded from here.

  Folder structure:

  data  
  ├── sequences  
  │   └── 00  
  │       ├── image_2  
  │       ├── Velodyne  
  │       └── poses.txt  
  └── ...
  

• The KITTI-360 DATASET The used data can be downloaded from here.

  Folder structure:

  data  
  ├── data_2d_raw    
  │   ├── 2013_05_28_drive_0002_sync  
  │   │   ├── image_00  
  │   │   └── ...  
  │   └── ...  
  ├── data_3d_raw  
  │   ├── 2013_05_28_drive_0002_sync  
  │   │   ├── image_00  
  │   │   └── ...  
  │   └── ...  
  └── data_poses  
      ├── 2013_05_28_drive_0002_sync  
      │   ├── image_00  
      │   └── ...  
      └── ...
  

💡 Visualization

Our image-to-point-cloud place recognition on the unseen KITTI test sequence.