gowthaminti/visodopcnn

Monocular Odometry using Optical flow and Deep Neural Networks

This project focuses on training and testing end to end Monocular visual odometry models using deep learning techniques. Visual odometry is a key component in robotics and autonomous systems, enabling the estimation of a camera or robot's motion and position based on visual input.

Overview

• The code uses deep convolutional neural networks (CNNs) and GPUs for optical flow estimation.
• The code leverages PTLFLow library to utilize the flow models
• Several models, such as Pcnn, Pcnn1, and Pcnn2, are available for testing and training.
• Training data is typically sourced from datasets like KITTI odometry and MPI-Sintel.
• The code includes both training and testing modes to help develop and evaluate visual odometry models.

Requirements

• Python 3.x
• PyTorch
• PIL (Python Imaging Library)
• tqdm
• Matplotlib
• NumPy
• PTLflow library

Getting Started

1. Clone the repository.

2. Ensure you have the required libraries installed.

   pip install -r requirements.txt
   

Download KITTI datahere

Download the KITTI images and only the right camera color images (image_02 folder) are required

the downloaded images will be placed at dataset/sequences/images_2/00/, dataset/images_2/01, ...

the images offered by KITTI is already rectified

Download the ground truth pose from KITTI Visual Odometry

you need to enter your email to request the pose data here

and place the ground truth pose at dataset/poses/

Downloand ptlflow fromhere

and export the path so that the flow models cna be accessed directly

```bash
export PYTHON_PACKAGE_PATH="/path/to/ptlflow/ptlflow"

The models can be run directly on Kitti dataset using optical flow models as ptllflow models

```bash
python flow_train.py --model Pcnn --mode train --datapath /path/to/your/dataset --bsize 8 --lr 0.0001 --train_iter 200  --checkpoint_path /path/to/checkpoints --save_res /path/to/results

python flow_train.py --model Pcnn --mode test --datapath /path/to/your/dataset --seq 00 --checkpoint_load /path/to/checkpoint/model.pth --save_res /path/to/results

BibTex

@misc{morimitsu2021ptlflow, author = {Henrique Morimitsu}, title = {PyTorch Lightning Optical Flow}, year = {2021}, publisher = {GitHub}, journal = {GitHub repository}, howpublished = {\url{https://github.com/hmorimitsu/ptlflow}} }