---
license: mit
language:
- en
metrics:
- f1
base_model:
- Qwen/Qwen2.5-VL-7B-Instruct
pipeline_tag: image-text-to-text
tags:
- building
- facade
- material
---

# BMAT — Building Facade Material Classifier

Fine-tuned from **Qwen2.5-VL-7B-Instruct** on 39,405 manually annotated facade images.

## Training Data

- **39,405** manually annotated facade images across **9 material categories**
- Categories: `brick` · `concrete` · `glass` · `metal` · `stone` · `stucco` · `tile` · `wood` · `other`
- Training labels available at [yhyJoy/BMAT](https://github.com/yhyJoy/BMAT/tree/main/data/label)

## Performance

| Metric | Score |
|--------|-------|
| Overall F1 | **0.91** |

## Usage
```python
from modelscope import Qwen2_5_VLForConditionalGeneration, AutoProcessor

model = Qwen2_5_VLForConditionalGeneration.from_pretrained("yinjoy30/BMAT/BMAT_finetune")
processor = AutoProcessor.from_pretrained("yinjoy30/BMAT/BMAT_finetune")
```

Prompt used during inference:
> *What is the facade material of the central building in this image? Please choose one of the following: [tile, brick, concrete, glass, metal, other, stone, stucco, wood] Only answer with the exact word from the list.*

## Base Model

[Qwen2.5-VL-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-VL-7B-Instruct)