Oops! Something went wrong
[next-mdx-remote-client] error compiling MDX: Could not parse expression with acorn More information: https://mdxjs.com/docs/troubleshooting-mdx
RoTA-Bench: Evaluating 3D Spatial Reasoning through Interactive Projection Alignment
Evaluates whether vision–language models can convert 2D observations into precise 3D control by introducing RoTA-Bench, an interactive projection-alignment benchmark that synthesizes silhouette-preserving 3D objects, guarantees solvable initializations, and provides 674 tasks with a stratified 100-task test set.
Introduction
Information
- Websitearxiv.org
- AuthorsQingchuan Ma, Yuexiao Ma, Xiang Meng, Yixuan Zou, Tianyu Xie, Wang Chen, Yuhui Zeng, Yuhang Wu, Xiawu Zheng, Rongrong Ji
- Published date2026/10/13
Categories
More Items
Answers spatial choice, scalar, and point questions by classifying among discretized candidates using pretrained attention projections from a Qwen3.5-4B backbone instead of autoregressive text generation. Trains on converted spatial QA data, adopts coarse-to-fine refinement and joint multi-stage supervision, and reports higher accuracy with several-fold faster inference than autoregressive baselines.
Rewrites text-to-image prompts with reinforcement learning that aligns policy updates to semantic spans and pools log-probability shifts within each span to weight token losses. Improves prompt engineering for frozen generators without extra renders (≈9.5% on GenEval, 7.6% on T2I-CompBench).
Provides a large corpus of runnable 3D programs and a multi-agent harness that generates, executes, measures, visually judges, and repairs executable 3D code to produce editable, verifiable 3D scenes. Includes ~269K runnable programs, 2.03M render–caption pairs, and an automated execution+repair pipeline.