2 docs tagged with "VLM"

An open-vocabulary visual grounding solution via the LocateAnything extension — based on the LocateAnything-3B vision-language model, it finds / counts arbitrary objects in a frame by natural language, with phrase grounding, text localization, GUI grounding, and pointing. Supports zero-shot detection and counting for classes a fixed detector doesn't know or ad-hoc needs.

A VLM document-understanding solution via the PaddleOCR-VL extension — high-accuracy multilingual OCR, table recognition (HTML), and key information extraction (KIE) through a remote PaddleOCR-VL 1.6 inference service. Suited to complex layouts, invoice/form structured extraction, and warped/photographed documents that paddle-ocr-v6 cannot handle well.