Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodology

TL;DR

This paper introduces TreeBench, a comprehensive benchmark for evaluating visual grounded reasoning, and TreeVGR, a training method that enhances reasoning accuracy and traceability in vision-language models.

Contribution

The paper presents a new benchmark, TreeBench, for holistic evaluation of visual reasoning, and a training paradigm, TreeVGR, that improves reasoning accuracy and traceability in models.

Findings

Models struggle with TreeBench, with none exceeding 60% accuracy.

TreeVGR improves performance on multiple benchmarks, including TreeBench.

Traceability enhances reasoning accuracy and model interpretability.

Abstract

Models like OpenAI-o3 pioneer visual grounded reasoning by dynamically referencing visual regions, just like human "thinking with images". However, no benchmark exists to evaluate these capabilities holistically. To bridge this gap, we propose TreeBench (Traceable Evidence Evaluation Benchmark), a diagnostic benchmark built on three principles: (1) focused visual perception of subtle targets in complex scenes, (2) traceable evidence via bounding box evaluation, and (3) second-order reasoning to test object interactions and spatial hierarchies beyond simple object localization. Prioritizing images with dense objects, we initially sample 1K high-quality images from SA-1B, and incorporate eight LMM experts to manually annotate questions, candidate options, and answers for each image. After three stages of quality control, TreeBench consists of 405 challenging visual question-answering pairs, even the most advanced models struggle with this benchmark, where none of them reach 60% accuracy, e.g., OpenAI-o3 scores only 54.87. Furthermore, we introduce TreeVGR (Traceable Evidence Enhanced Visual Grounded Reasoning), a training paradigm to supervise localization and reasoning jointly with reinforcement learning, enabling accurate localizations and explainable reasoning pathways. Initialized from Qwen2.5-VL-7B, it improves V* Bench (+16.8), MME-RealWorld (+12.6), and TreeBench (+13.4), proving traceability is key to advancing vision-grounded reasoning. The code is available at https://github.com/Haochen-Wang409/TreeVGR.