README.md

---
title: MAOAM
emoji: 🎨
colorFrom: yellow
colorTo: pink
sdk: gradio
sdk_version: "5.29.0"
app_file: app.py
short_description: Object and Material Selection VLM
python_version: "3.10"
startup_duration_timeout: "600s"
---

# MAOAM: Unified Object and Material Selection with Vision-Language Models

This demo showcases **MAOAM** (Mask Any Object And Material), a unified selection framework that enables precise object- and material-level segmentation across both text- and click-based interactions.

## How to use

1. **Upload an image** — any RGB photo works.
2. **Choose a selection mode**:
   - **Material: click** — Place star markers on the material you want to segment.
   - **Material: text** — Describe the material in words (e.g., "shiny chrome metal").
   - **Object: text** — Name an object (e.g., "the chair").
3. **Click Submit** — The model produces a segmentation mask overlaid on your image.

## Model

This Space uses the **MAOAM-Sa2VA** variant, based on Qwen2.5-VL-7B + SAM2 Hiera-L, fine-tuned for unified object and material selection.

- Paper: [MAOAM: Unified Object and Material Selection with Vision-Language Models](https://arxiv.org/abs/2606.04880)
- Code: [github.com/adobe-research/obj-and-mat-selection](https://github.com/adobe-research/obj-and-mat-selection)
- Weights: [jpark677/maoam_ckpts](https://huggingface.co/jpark677/maoam_ckpts)