Optimizing Food-to-3D Model Automated Pipeline

Computer Vision & 3D Generative AI Engineer – Automated Food-to-3D Model Pipeline, Project Overview

We are building an automated pipeline that converts user-uploaded photos of food and restaurant dishes into clean, photorealistic 3D assets (.glb / .gltf).

Currently, feeding raw user photos into 3D reconstruction APIs (Tripo3D ) causes high-frequency textures (like rice, grains, or sauces) to turn into distorted, lumpy geometric artifacts ("stones" / "slugs"), while moving cutlery and table backgrounds break multiview alignment.

We are looking for an experienced developer with expertise in Computer Vision pre-processing and Generative 3D APIs to design, build, and optimize an automated end-to-end processing pipeline.

Paid / Trial Milestone: Sample Verification Task
Trial Task Requirement (Images Attached):
Attached to this post are 4 photos of a bowl of fried rice taken from different angles.

To be considered for this project, you must demonstrate how you solve the common reconstruction issues with this specific test case:

Process the provided images to clean the background, isolate the bowl, and eliminate the shifting cutlery.

Generate a preview 3D model (.glb or interactive web viewer link, or short screen recording of the wireframe and textured model).

The Acceptance Criteria: The food inside the bowl must generate as a clean, smooth surface (not a chaotic cluster of bumpy "stones" or melted geometry), with the fried rice texture mapped realistically across it.