README.md

---
title: GroundingPI Perception Studio
emoji: 🎯
colorFrom: purple
colorTo: green
sdk: gradio
sdk_version: 5.49.1
python_version: 3.12.12
app_file: space_app.py
pinned: false
models:
  - GroundingPI/GroundingPI
---

# GroundingPI · Perception Studio

The original English demo, with its purple and lime palette, fonts, layout, nine task routes and 22 image examples. Supports image upload, canvas zoom, target focus, original-pixel detail, downloads and structured outputs. Video remains disabled.

The model is `GroundingPI/GroundingPI`, pinned at revision `127b0e1c54c68f24bf769a80067521b815b695dd`. Native processors, chat templates and GAM coordinates are preserved. Chinese OCR labels use the bundled Noto Sans CJK font.

## Local preview

```bash
python3 -m venv .venv
.venv/bin/python -m pip install -r requirements.txt
GROUNDING_BACKEND=http DEMO_PORT=7862 .venv/bin/python space_app.py
```

The preview uses `http://127.0.0.1:8000/v1` by default. Without a model service, it reports inference offline and allows browsing examples. It does not substitute stored predictions.

## ZeroGPU

Deploy this directory to the existing `FanLR/GroundingPI-4B` Gradio Space. The Space and GPU backend belong to FanLR; model weights come from the public GroundingPI account repository. `space_app.py` replaces the old Space model variables with this pinned release. Do not configure both a local model path and a remote model repository.

## Verification

```bash
.venv/bin/python -m unittest discover -s tests -v
.venv/bin/python scripts/verify_ui.py
.venv/bin/python scripts/build_bundle.py /absolute/path/studio.zip
```

Local tests and CPU package validation are recorded separately from live GPU inference. Images, fonts and model weights retain their source licenses; see `assets/SOURCES.md` and the model repository.