# A Cross-Model VLM-Judge Protocol for Single-Image 3D Mesh Quality (and Why Cheap Proxies Fall Short)

arXiv: https://arxiv.org/abs/2606.18451 · HTML: https://mlsanity.com/paper/2606.18451/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 29 references that are in the map.
- First posted on arXiv 16 Jun 2026.

## Builds on (most cited first, 22 of 22)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [InternVL3: Exploring Advanced Training and Test-Time Recipes for Open-Source Multimodal Models](https://mlsanity.com/paper/2504.10479/) (2025)
- [Large Language Models are not Fair Evaluators](https://mlsanity.com/paper/2305.17926/) (2023)
- [Google Scanned Objects: A High-Quality Dataset of 3D Scanned Household Items](https://mlsanity.com/paper/2204.11918/) (2022)
- [TripoSR: Fast 3D Object Reconstruction from a Single Image](https://mlsanity.com/paper/2403.02151/) (2024)
- [Unique3D: High-Quality and Efficient 3D Mesh Generation from a Single Image](https://mlsanity.com/paper/2405.20343/) (2024)
- [SF3D: Stable Fast 3D Mesh Reconstruction with UV-unwrapping and Illumination Disentanglement](https://mlsanity.com/paper/2408.00653/) (2024)
- [DreamDPO: Aligning Text-to-3D Generation with Human Preferences via Direct Preference Optimization](https://mlsanity.com/paper/2502.04370/) (2025)
- [DSO: Aligning 3D Generators with Simulation Feedback for Physical Soundness](https://mlsanity.com/paper/2503.22677/) (2025)
- [ShapeR: Robust Conditional 3D Shape Generation from Casual Captures](https://mlsanity.com/paper/2601.11514/) (2026)
- [Objaverse++: Curated 3D Object Dataset with Quality Annotations](https://mlsanity.com/paper/2504.07334/) (2025)
- [Nabla-R2D3: Effective and Efficient 3D Diffusion Alignment with 2D Rewards](https://mlsanity.com/paper/2506.15684/) (2025)
- [MVGBench: Comprehensive Benchmark for Multi-view Generation Models](https://mlsanity.com/paper/2507.00006/) (2025)
- [ImagenWorld: Stress-Testing Image Generation Models with Explainable Human Evaluation on Open-ended Real-World Tasks](https://mlsanity.com/paper/2603.27862/) (2026)
- [ManiTwin: Scaling Data-Generation-Ready Digital Object Dataset to 100K](https://mlsanity.com/paper/2603.16866/) (2026)
- [MJ1: Multimodal Judgment via Grounded Verification](https://mlsanity.com/paper/2603.07990/) (2026)
- [Low-Rank Adaptation of Neural Fields](https://mlsanity.com/paper/2504.15933/) (2025)
- [Rethinking Metrics and Diffusion Architecture for 3D Point Cloud Generation](https://mlsanity.com/paper/2511.05308/) (2025)
- [Realiz3D: 3D Generation Made Photorealistic via Domain-Aware Learning](https://mlsanity.com/paper/2605.13852/) (2026)
- [Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework](https://mlsanity.com/paper/2509.02474/) (2025)
- [Memorization in 3D Shape Generation: An Empirical Study](https://mlsanity.com/paper/2512.23628/) (2025)

## Built on by (most cited first, 2 of 2)

- [Judging to Improve: A De-biased VLM-as-3D-Judge Protocol for Single-Image 3D Generation](https://mlsanity.com/paper/2606.20364/) (2026)
- [Frozen Scenes, Shifting Winners: Configuration Fragility in Text-to-3D Evaluation](https://mlsanity.com/paper/2610.00447/) (2026)

## Abstract (arXiv)

Single-image-to-3D generators are improving quickly, but there is no agreed, human-free way to tell whether one generated mesh is better than another. Practitioners commonly rely on cheap automatic proxies (render-space CLIP similarity and mesh geometry-validity statistics), yet how well these track perceived quality is unestablished. We make two contributions. First, we propose and validate a reproducible VLM-judge evaluation protocol: a fixed 24-view headless render rig, two independent vision-language judge families, and a mandatory position-bias correction that queries both presentation orders and keeps only order-consistent verdicts. The two judge families agree substantially with each other (Cohen's kappa = 0.66), well above the chance-agreement floor. Second, using this protocol as the reference, we show the cheap proxies do not substitute for it. Geometry validity is only a weak signal on average (because, as we show, it is bimodal) and stays below our pre-registered target, while render-CLIP is at chance. A learned Bradley-Terry head collapses onto a single manifoldness statistic (giving render-CLIP a negative weight) and matches geometry-only exactly, so learning the feature weights buys nothing. The proxy is also bimodal: it is significantly above chance on contrasts with visible geometric defects but at chance on ambiguous contrasts, consistent with geometry validity tracking the judge only when the defect is visually salient. We therefore recommend the VLM-judge protocol as a reliable, reproducible evaluator under the conditions tested (two feed-forward generators on Google Scanned Objects, with a face-drop degradation regime) and advise against geometry/CLIP proxies as optimization targets.