Spaces:
Running on Zero
Running on Zero
| title: Seg2Click | |
| emoji: 🖱️ | |
| colorFrom: blue | |
| colorTo: green | |
| sdk: gradio | |
| sdk_version: 6.30.0 | |
| app_file: app.py | |
| python_version: "3.12" | |
| short_description: Pixel-level action masks for GUI visual grounding | |
| startup_duration_timeout: 30m | |
| suggested_hardware: zero-a10g | |
| # Seg2Click | |
| [Seg2Click](https://github.com/jcwang0602/Seg2Click) turns a GUI screenshot and a natural-language | |
| instruction into a pixel-level executable action mask and click point. This Space runs the lightweight public | |
| [Seg2Click-Qwen3.5-0.8B](https://huggingface.co/jcwang0602/Seg2Click-Qwen3.5-0.8B) checkpoint with the official | |
| inference path. | |
| Seg2Click-0.8B achieves **87.4 / 89.9 / 44.3** Hit@1 on ScreenSpot / ScreenSpot-v2 / ScreenSpot-Pro. | |
| The example screenshots come from [ScreenSpot-Pro](https://huggingface.co/datasets/likaixin/ScreenSpot-Pro) | |
| and are evaluated with fresh model inference whenever an example is clicked. | |
| Read the [paper](https://arxiv.org/abs/2610.16234) for method and evaluation details. | |