---
license: apache-2.0
datasets:
- COCO
- ReasonSeg
- CountBench
language:
- en
metrics:
- accuracy
base_model:
- Qwen2.5-VL
pipeline_tag: image-text-to-text
library_name: transformers
---

# VisionReasoner-7B

[Paper](https://huggingface.co/papers/2505.12081)

Code: [https://github.com/dvlab-research/VisionReasoner](https://github.com/dvlab-research/VisionReasoner)

Project page: [https://github.com/dvlab-research/VisionReasoner](https://github.com/dvlab-research/VisionReasoner)

## Description

This is a VisionReasoner-7B model. It introduces a decoupled architecture consisting of a reasoning model and a segmentation model. The reasoning model interprets user intentions, generates explicit reasoning chains, and produces positional prompts, which are subsequently used by the segmentation model to generate pixel-level masks.

## Usage

```python
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

# load model
model = AutoModelForCausalLM.from_pretrained("Ricky06662/VisionReasoner-7B")
tokenizer = AutoTokenizer.from_pretrained("Ricky06662/VisionReasoner-7B")
```