xby-pic

Processing images inside a conversational AI pipeline usually means juggling dozens of vision APIs that each speak a different dialect of auth, request shape, and response schema. xby-pic collapses more than thirty image-understanding tasks into a single DeepSeek Harness plugin, so you call one entry point from chat and get everything from OCR to instance segmentation without writing any glue code.

Capabilities span three tiers. The first is structured information extraction: standard and high-accuracy OCR, handwriting recognition, and structured reads of driver licenses, ID cards, passports, HK/MO/TW travel documents, bank cards (with Luhn validation), business licenses, and vehicle registration certificates. Every endpoint accepts a URL, a base64 string, or a local file path. The second tier targets industry-specific detection: fire, smoke, pedestrians, vehicles, e-bikes, safety helmets, reflective vests, smoking, and gestures in surveillance and traffic scenarios. The third tier is general-purpose vision: ImageNet-1000 classification, COCO-80 object detection and instance segmentation, oriented (rotated) bounding-box detection across 15 ADE20K-style classes, 17-keypoint human pose estimation, and an open-vocabulary recognition model covering over fifty thousand object categories, plus fine-grained recognition of insects, plants, birds, dishes, and pet emotions.

It is built for developers extending DSH chat pipelines with batch image processing, engineers assembling automated document-verification or safety-monitoring workflows, and business teams that need a single toolkit covering certificate review, security monitoring, and general visual classification behind one API key from the XiaoBenYang platform.