README.md

---
license: apache-2.0
base_model: Qwen/Qwen3-VL-32B-Instruct
tags:
  - spatial-reasoning
  - multi-hop
  - grounding
  - vision-language
  - qwen3-vl
  - GRPO
language:
  - en
datasets:
  - etri-vilab/MultihopSpatial
pipeline_tag: image-text-to-text
---

# [ECCV 2026] MultiHopSpatial-Qwen3-VL-32B-Instruct

This model is [Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct) post-trained on [MultihopSpatial-Train](https://huggingface.co/datasets/etri-vilab/MultihopSpatial) using **GRPO (Group Relative Policy Optimization)** for multi-hop spatial reasoning.

<p align="center">
  <a href="https://youngwanlee.github.io/multihopspatial"><b>Project Page</b></a> |
  <a href="https://arxiv.org/abs/2603.18892"><b>Paper</b></a> |
  <a href="https://huggingface.co/datasets/etri-vilab/MultihopSpatial"><b>Dataset</b></a>
</p>

## Model Zoo

| Model | Params | HF Link |
|---|---|---|
| MultiHopSpatial-Qwen3-VL-4B-Instruct | 4B | [🤗 etri-vilab/MultiHopSpatial-Qwen3-VL-4B-Instruct](https://huggingface.co/etri-vilab/MultiHopSpatial-Qwen3-VL-4B-Instruct) |
| MultiHopSpatial-Qwen3-VL-8B-Instruct | 8B | [🤗 etri-vilab/MultiHopSpatial-Qwen3-VL-8B-Instruct](https://huggingface.co/etri-vilab/MultiHopSpatial-Qwen3-VL-8B-Instruct) |
| MultiHopSpatial-Qwen3-VL-32B-Instruct | 32B | [🤗 etri-vilab/MultiHopSpatial-Qwen3-VL-32B-Instruct](https://huggingface.co/etri-vilab/MultiHopSpatial-Qwen3-VL-32B-Instruct) |

## Model Details

| | |
|---|---|
| **Base Model** | [Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct) |
| **Architecture** | Qwen3VLForConditionalGeneration |
| **Training Method** | GRPO (Group Relative Policy Optimization) |
| **Training Data** | [MultihopSpatial-Train](https://huggingface.co/datasets/etri-vilab/MultihopSpatial) (6,791 samples) |
| **Precision** | bfloat16 |

## Results

<p align="center">
  <img src="result.png" width="100%" alt="Training corpus comparison across model scales">
</p>

## Usage

This model shares the same architecture and usage as [Qwen3-VL-32B-Instruct](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct). Please refer to the [official Qwen3-VL documentation](https://huggingface.co/Qwen/Qwen3-VL-32B-Instruct) for detailed usage instructions.

### Quick Start

```python
from transformers import Qwen3VLForConditionalGeneration, AutoProcessor
from qwen_vl_utils import process_vision_info

model = Qwen3VLForConditionalGeneration.from_pretrained(
    "etri-vilab/MultiHopSpatial-Qwen3-VL-32B-Instruct",
    torch_dtype="auto",
    device_map="auto",
)
processor = AutoProcessor.from_pretrained("etri-vilab/MultiHopSpatial-Qwen3-VL-32B-Instruct")

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "image": "your_image.jpg"},
            {"type": "text", "text": "From the perspective of the person wearing a red shirt, which object is on their left? (a) chair (b) table (c) lamp (d) bookshelf. And provide the bounding box coordinate of the region related to your answer."},
        ],
    }
]

text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
image_inputs, video_inputs = process_vision_info(messages)
inputs = processor(
    text=[text],
    images=image_inputs,
    videos=video_inputs,
    padding=True,
    return_tensors="pt",
).to(model.device)

generated_ids = model.generate(**inputs, max_new_tokens=2048)
output_text = processor.batch_decode(
    generated_ids[:, inputs.input_ids.shape[1]:], skip_special_tokens=True
)
print(output_text[0])
```

## Citation

```bibtex
@inproceedings{lee2026multihopspatial,
  title={MultihopSpatial: Multi-hop Compositional Spatial Reasoning Benchmark for Vision-Language Models},
  author={Lee, Youngwan and Jang, Soojin and Cho, Yoorhim and Lee, Seunghwan and Lee, Yong-Ju and Hwang, Sung Ju},
  booktitle={European Conference on Computer Vision (ECCV)},
  year={2026}
}
```