CurrI2P

Get Started

Installation and Data Preparation

step 1. Prepare CurrI2P repo by.

git clone https://github.com/lin-liwei/CurrI2P.git
cd CurrI2P

step 2. Please prepare environment.

We implement our method on two baselines, and their environments are the same as their baselines. Therefore, you can refer to:

• VP2P
• CorrI2P

The inference code was tested on:

• Ubuntu 16.04 LTS, Python 3.7, Pytorch 1.7.1, CUDA 9.2, GeForce RTX 2080Ti (pip, Conda)

conda create -n CurrI2P python=3.7
conda activate CurrI2P
pip install -r requirements.txt

step 3. Download data.

We provide the inference data from the KITTI dataset (sequences 9-10) in here to help you quickly get started with evaluating CurrI2P (VP2P). The data tree should be arranged as:

kitti/
├── calib
    └── 09
    └── 10
├── data_odometry_color
    └── sequences 
        ├── 09
        ├── 10
├── data_odometry_velodyne
    └── sequences 
        ├── 09
        ├── 10

Inference

• Checkpoints. We provide the pre-trained checkpoint of CurrI2P(VP2P) in here.
• Scripts. After prepraring the code and checkpoints, you can simply evaluate CurrI2P(VP2P) by runing:

python main.py

Train

To train CurrI2P, begin by preparing the training data as outlined in VP2P or CorrI2P. Additional instructions for training CurrI2P will be released soon.

Acknowledgements

This implementation is based on / inspired by:

• VP2P
• CorrI2P
• DeepI2P