Update README with OMR project description and Starry features

README.md

@@ -9,17 +9,30 @@ license: bsd
 app_port: 7860
 ---
 
-# STARRY — Sheet Music Recognition
+# STARRY — Optical Music Recognition
 
-Online sheet music recognition and editing platform.
+Starry is a demo application for an end-to-end **Optical Music Recognition (OMR)** system. It transforms scanned sheet music images into structured digital music notation through a multi-stage ML pipeline, and provides an interactive platform for reviewing, editing, and managing the recognition results.
 
-## Features
+## What Makes Starry Different
 
-- Upload and manage music score images
-- Automatic OMR (Optical Music Recognition) processing
-- Score editing and annotation
-- Music set management
+- **Multi-stage ML pipeline** — Seven specialized models work in sequence: layout detection, staff gauge prediction, foreground/background mask separation, symbol semantic recognition, text location detection, OCR, and bracket recognition. Each model focuses on a specific subtask, enabling high overall accuracy.
+- **Automatic regulation** — An ONNX-based solver analyzes recognized measures for timing consistency and automatically corrects quantization errors, reducing the need for manual intervention.
+- **Measure-level quality tracking** — The system identifies and flags problematic measures for human review, with an annotation workflow that tracks corrections per-measure.
+- **Real-time progress** — WebSocket streaming keeps the UI updated during long-running recognition and regulation tasks.
+- **Score collection management** — Organize recognized scores into music sets with tagging and categorization, useful for building OMR datasets or curating repertoire.
+
+## Architecture
+
+```
+Frontend (React)
+  ↓ nginx (port 7860)
+  ├─ /api/score → cluster-server (NestJS)
+  └─ /api/*     → omr-service (Fastify)
+                    ├─ async task worker
+                    ├─ PostgreSQL
+                    └─ ZMQ → Python ML services (×7)
+```
 
 ## Note
 
-This is a lightweight deployment. ML prediction services (layout, gauge, mask, semantic, OCR) are not included in this Space.
+This is a lightweight demo deployment on HuggingFace Spaces. The ML prediction services (layout, gauge, mask, semantic, text, OCR, brackets) are **not included** in this Space — it is intended for showcasing the review and editing workflow with pre-computed results.