OCR Service

The OCR service extracts player name, run time, and map code from Overwatch parkour screenshots. It is implemented in the genjishimada-ocr repository and is consumed by the API during completion submission flows.

Purpose

The OCR service processes parkour screenshots and returns structured data including the player's name, run time (seconds), and map code. It also returns raw extracted text for debugging.

Key features

• Multi-language OCR with prewarmed models (English, Chinese, Korean, Japanese).
• Script-aware name selection to choose the most plausible player name.
• Robust time parsing with error correction for common OCR misreads.
• Flexible map code extraction using explicit and generic patterns.

API endpoints

Method & Path	Description
GET /ping	Returns { "ok": true, "models": [...] } and readiness info.
POST /extract	Accepts { "image_b64": "data:image/png;base64,..." } and returns extracted data.

Interactive documentation is available at /docs (Swagger UI) and /redoc on the OCR service.

Deployment & quickstart

The OCR service runs independently and is not deployed by this repo. See the genjishimada-ocr repository for Docker and local Python setup details.