FoodRiddle-MetaFood-CVPR2024

Overview

This project is developed for the CVPR MetaFood 2024 Workshop Challenge.

The goal of our project is to accurately model and analyze food through 3D reconstruction, volume estimation, and mesh reconstruction.

Team FoodRiddle

Team Members

• Yawei Jueluo: [email protected]
• Chengyu Shi: [email protected]
• Pengyu Wang: [email protected]

Features

• 3D Reconstruction: High-precision 3D reconstruction using state-of-the-art computer vision techniques.
• Volume Estimation: Accurate volume estimation through advanced algorithms.
• Mesh Reconstruction: High-quality mesh reconstruction addressing the problem of incomplete edges.

Dataset

We used the MetaFood 2024 Challenge Dataset for training and evaluating our models.

Data Preparation

1. Download the dataset from the provided link.
2. Preprocess the data and run SFM using the provided scripts:

   python core/sfm_pipeline.py --dataset_root /path/to/data

3. Run 2D gaussian splatting with result
4. Run demo to see the final results with

   python core/mesh_tool.py

   and

   python core/post_process.py

Installation and Dependencies

Environment Setup

Ensure you have the following environment:

• Python version: 3.8+
• CUDA version: 11.0+ (if using GPU)

Dependencies

Install all dependencies using the following command:

pip install -r requirements.txt