TreeVGR-7B-CI Open-source Visual Localization Inference Model - Provides Accurate Locations and Explainable Inference Paths

Treevgr 7B CI

Developed by HaochenWang

TreeVGR-7B is a visual positioning reasoning model with traceable evidence enhancement. It combines supervised positioning and reasoning through reinforcement learning to achieve accurate positioning and interpretable reasoning paths.

Text-to-Image

Transformers

Open Source License:Apache-2.0 #Visual positioning reasoning #Traceable evidence enhancement #Second-order reasoning ability

Downloads 115

Release Time : 7/3/2025

Model Overview

TreeVGR-7B is an advanced open-source visual positioning reasoning model, initialized based on Qwen2.5-VL-7B, and performs excellently in multiple benchmark tests.

Model Features

Traceable evidence enhancement

Combines supervised positioning and reasoning through reinforcement learning to achieve accurate positioning and interpretable reasoning paths.

Complex scene processing

Capable of handling complex scenes with dense objects and focusing on visual perception of subtle targets.

Second-order reasoning ability

Tests object interactions and spatial hierarchies, rather than just simple object positioning.

Model Capabilities

Visual positioning reasoning

Complex scene analysis

Second-order reasoning

Interpretable reasoning path

Use Cases

Visual question answering

TreeBench evaluation

Conduct visual question answering evaluation on TreeBench to test the model's visual perception and reasoning abilities.

Achieved an accuracy of 49.38% and a Mean IoU of 43.3 on TreeBench.

Visual positioning

V* Bench evaluation

Conduct visual positioning evaluation on V* Bench to test the model's positioning ability.

The performance is improved by 16.8%.

🚀 TreeVGR-7B: Traceable Evidence Enhanced Visual Grounded Reasoning Model

This repository hosts the TreeVGR-7B model, a cutting - edge open - source visual grounded reasoning model. It addresses the challenges in visual reasoning by providing a novel approach and evaluation benchmark.

🚀 Quick Start

The TreeVGR-7B model offers a new way to approach visual grounded reasoning. This README will guide you through its installation, usage, and more.

✨ Features

Novel Benchmark: TreeBench is a diagnostic benchmark designed to comprehensively evaluate visual grounded reasoning capabilities, with a focus on subtle target perception, traceable evidence, and second - order reasoning.
Training Paradigm: TreeVGR is a training paradigm that jointly supervises localization and reasoning using reinforcement learning, improving performance on multiple benchmarks.

📦 Installation

pip3 install -r requirements.txt
pip3 install flash-attn --no-build-isolation -v

💻 Usage Examples

Basic Usage

First, clone this repo:

git clone https://github.com/Haochen-Wang409/TreeVGR
cd TreeVGR

Then, run the inference script:

python3 inference_treebench.py

This should give:

Perception/Attributes 18/29=62.07
Perception/Material 7/13=53.85
Perception/Physical State 19/23=82.61
Perception/Object Retrieval 10/16=62.5
Perception/OCR 42/68=61.76
Reasoning/Perspective Transform 19/85=22.35
Reasoning/Ordering 20/57=35.09
Reasoning/Contact and Occlusion 25/41=60.98
Reasoning/Spatial Containment 20/29=68.97
Reasoning/Comparison 20/44=45.45
==> Overall 200/405=49.38
==> Mean IoU: 43.3

Note that this result is slightly different from the paper, as a more comprehensive evaluation was done using VLMEvalKit.

📚 Documentation

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.

TreeBench Overview

News

[2025/07/11] 🔥🔥🔥 TreeBench and TreeVGR have been supported by VLMEvalKit! 🔥🔥🔥
[2025/07/11] 🔥 TreeBench and TreeVGR have been released.

Hugging Face Resources

Property	Details
Benchmark	TreeBench
Checkpoints	TreeVGR-7B, TreeVGR-7B-CI
Training Datasets	TreeVGR-RL-37K, TreeVGR-SFT-35K

Citation

If you find this work useful for your research and applications, please cite using this BibTeX:

@article{wang2025traceable,
  title={Traceable Evidence Enhanced Visual Grounded Reasoning: Evaluation and Methodology},
  author={Haochen Wang and Xiangtai Li and Zilong Huang and Anran Wang and Jiacong Wang and Tao Zhang and Jiani Zheng and Sule Bai and Zijian Kang and Jiashi Feng and Zhuochen Wang and Zhaoxiang Zhang},
  journal={arXiv preprint arXiv:2507.07999},
  year={2025}
}