Neural Mesh Simplification

This repository contains an implementation of the paper "Neural Mesh Simplification" by Potamias et al. (CVPR 2022). The project aims to provide a fast, learnable method for mesh simplification that generates simplified meshes in real-time.

Research, methodology introduced in the Neural Mesh Simplification paper, with the updated info shared in supplementary material.

This implementation could not have been done without the use of an LLM, specifically Claude Sonnet 3.5 by Anthropic. It was useful to create a project, upload the PDF of the papers there and use the custom instructions in llm_instructions.txt. To steer the model, a copy of the file structure (which it helped create early on) is also useful. This can be created with the command tree -F -I '*.ply|*.obj|*__pycache__*' > file_structure.txt in the root directory of the project.

It is also useful to keep an updated copy of the main components of the code-base in the LLM project. This can be done with the following command, and uploading the file to the project in Claude:

find losses metrics models tests utils trainer \
    \( -type d -name "*__pycache__*" -o -name ".DS_Store" \) -prune -o -type f -print \
    | while IFS= read -r filepath; do
        echo "====== $filepath start ======" >> combined_output.txt
        cat "$filepath" >> combined_output.txt
        echo "====== end of $filepath ======" >> combined_output.txt
    done

Overview

Neural Mesh Simplification is a novel approach to reduce the resolution of 3D meshes while preserving their appearance. Unlike traditional simplification methods that collapse edges in a greedy iterative manner, this method simplifies a given mesh in one pass using deep learning techniques.

The method consists of three main steps:

Sampling a subset of input vertices using a sophisticated extension of random sampling.
Training a sparse attention network to propose candidate triangles based on the edge connectivity of sampled vertices.
Using a classification network to estimate the probability that a candidate triangle will be included in the final mesh.

Features

Fast and scalable mesh simplification
One-pass simplification process
Preservation of mesh appearance
Lightweight and differentiable implementation
Suitable for integration into learnable pipelines

Installation

git clone https://github.com/martinnormark/neural-mesh-simplification.git
cd neural-mesh-simplification
pip install -r requirements.txt

Usage

from neural_mesh_simplifier import NeuralMeshSimplifier

# Initialize the simplifier
simplifier = NeuralMeshSimplifier()

# Load a mesh
original_mesh = load_mesh("path/to/your/mesh.obj")

# Simplify the mesh
simplified_mesh = simplifier.simplify(original_mesh, target_faces=1000)

# Save the simplified mesh
save_mesh(simplified_mesh, "path/to/simplified_mesh.obj")

Training

To train the model on your own dataset:

python ./scripts/train.py --data_path /path/to/your/dataset --epochs 100 --batch_size 32

Evaluation

To evaluate the model on a test set:

python ./scripts/evaluate.py --model_path /path/to/saved/model --test_data /path/to/test/set

Citation

If you use this code in your research, please cite the original paper:

@InProceedings{Potamias_2022_CVPR,
    author    = {Potamias, Rolandos Alexandros and Ploumpis, Stylianos and Zafeiriou, Stefanos},
    title     = {Neural Mesh Simplification},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
    pages     = {18583-18592}
}

Contributing

Contributions are welcome to improve this implementation. Please feel free to submit issues and pull requests.

License

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

martinnormark / neural-mesh-simplification

readme