dreamers-laboratory/image-to-3d-pipeline

What it solves

It provides a standardized pipeline to compare different open-source image-to-3D reconstruction models. Because there is no single clear winner among available models, this project allows users to run multiple models on the same input images and evaluate their outputs using a consistent inspection protocol.

How it works

The pipeline follows a four-stage process:

1. Preprocess: Creates transparent object-only masks from source renders.
2. Reconstruct: Uses various third-party models (such as TRELLIS, TripoSR, stable-fast-3d, and Hunyuan3D-2) to generate 3D meshes.
3. Evaluate: Renders each resulting mesh through a fixed Blender orbit to create contact sheets for visual scoring.
4. Explore: Loads the final chosen mesh into a WebGL-based browser explorer for interactive viewing.

Who it’s for

Developers and researchers interested in 3D reconstruction who want to objectively compare the quality of different AI models to determine which one best suits their specific assets.

Highlights

• Comparative Framework: Integrates multiple state-of-the-art reconstruction tools into one workflow.
• Standardized Evaluation: Uses a fixed Blender inspection protocol to ensure fair comparison between models.
• WebGL Integration: Includes a browser-based explorer for navigating the reconstructed 3D objects.
• Multi-view Support: Handles multi-view reasoning by starting from a sheet of multiple viewpoints of an object.