An open model from the WeChat vision team handles detection, counting, OCR and pixel masking without bolting on…