Flow-Based Point Cloud Prediction

Overview

Flow-Based Point Cloud Prediction is a deep learning-based framework that predicts future point clouds given an initial point cloud. The model leverages flow matching, which learns the transformation (flow) from one point cloud to another, enabling accurate predictions of future states.

Features

• Flow Matching Learning: The model learns the flow between two consecutive point clouds.
• Future State Prediction: Given an initial point cloud, the model predicts how it evolves over time.
• Pretrained Weights: The experiments/ section includes pretrained weights for quick evaluation.
• Modular Implementation: Well-structured scripts for training, evaluation, and optimization.

Codebase Structure

Flow-based-point-cloud-prediction/
│── experiments/        # Contains pretrained weights and experiment logs
│── datasets.py         # Data loading and preprocessing functions
│── main.py             # Entry point to train and evaluate the model
│── models.py           # Model architecture for flow matching
│── optimisation.py     # Optimization algorithms and loss functions

Installation

Clone the repository and install the required dependencies:

git clone https://github.com/maheshreddy24/Flow-based-point-cloud-prediction.git
cd Flow-based-point-cloud-prediction
pip install -r requirements.txt

Usage

Running the Model

To start the training or prediction process, simply run:

python main.py

This script handles data loading, training, and evaluation.

Using Pretrained Weights

To use pretrained weights for evaluation, modify the configuration in experiments/ and load the corresponding checkpoint.

Future Work

• Enhancing flow estimation accuracy.
• Extending support for more point cloud datasets.
• Improving efficiency in large-scale point cloud processing.

License

This project is licensed under the MIT License. See the LICENSE file for details.