Cursor-follow OCR benchmark for macOS. Captures a square region around the mouse, runs OCR, shows a live preview of the exact image sent to OCR plus latency / FPS / memory stats. Handy for picking an OCR backend for on-screen text.
| Script | Backend |
|---|---|
apple_vision_cursor_ocr_benchmark.py |
Apple Vision (on-device, ANE/GPU) |
ppocrv5_server.py |
PaddleOCR PP-OCRv5, text only |
ppocrv5_cursor_profiler.py |
PaddleOCR PP-OCRv5, text + bounding boxes/polygons overlaid |
Live controls: change capture box size, change OCR input scale, nudge OCR
frequency. q/ESC to quit.
pip install -r requirements.txtApple Vision path needs only the pyobjc frameworks. Paddle path needs
paddleocr + paddlepaddle (CPU-only on stock macOS — the profiler measures
the full Python/Paddle pipeline, not raw model latency).
Screen Recording for the terminal/app you run this from.