3D-DefectBench: A Controlled Factorial Study of Vision-Language Model Evaluation Pipelines for Fine-Grained 3D Generation Defects

2026-07-12

AI Summary

This work addresses the lack of systematic evaluation frameworks for assessing the end-to-end reliability of existing 3D generation models and their ability to diagnose fine-grained defects. The authors propose 3D-DefectBench, a novel benchmark that, for the first time, disentangles the influences of vision-language models (VLMs), camera protocols, visual inputs, and prompt templates through carefully balanced factors. It establishes a controlled evaluation suite encompassing nine binary defect categories, including geometry, texture, and instruction following. Based on 3.2 million human-annotated defect judgments, the study reveals that model choice is the primary factor affecting agreement with human assessments. A lightweight six-view RGB input strategy achieves performance comparable to dense multi-view or depth-augmented alternatives; however, even the best VLMs significantly underperform expert annotators, particularly exhibiting degraded texture consistency under noisy labeling conditions.

Abstract

Automated evaluation is essential for scaling generative 3D systems, where exhaustive human review is costly and slow. However, the reliability of an automated judge depends on the entire evaluation pipeline, not only the underlying vision-language model (VLM), but also how assets are rendered, what visual evidence is provided, how the task is specified, and how human reference labels are constructed. We introduce 3D-DefectBench, a benchmark and framework for systematic analysis of VLM-based 3D defect detection pipelines. It complements holistic ratings and pairwise preferences with nine fine-grained binary defects spanning geometry, texture, and prompt adherence, providing actionable diagnostics for generator development and judge evaluation. Using a balanced factorial design, we vary four pipeline factors, VLM, camera protocol, visual input, and prompt schema, across 84 inference designs and approximately 3.2 million scored defect decisions, followed by staged validation on a broader set of frontier models. Model choice is the largest determinant of agreement with human labels, but the remaining factors also affect performance, interact with model selection, and can change the best configuration. Within the evaluated design space, a compact six-view RGB protocol performs comparably to denser multi-view settings and inputs augmented with depth or surface normals, making it a strong cost-effective default. Under this standardized pipeline, the best of 12 VLM judges still lag behind trained human labelers, while texture agreement drops sharply when expert-consensus labels are replaced by noisier silver labels. These findings show that automated judges should be evaluated as complete pipelines and calibrated across human reference regimes, rather than benchmarked only as standalone models. We release labels, prompts, predictions, and Croissant metadata on Hugging Face.

Problem

• 3D generation defects
• vision-language models
• automated evaluation
• evaluation pipeline
• fine-grained defect detection

Innovation

• 3D-DefectBench
• vision-language model
• factorial evaluation
• fine-grained defect detection
• automated 3D assessment