Show HN: Browser-based PDF form fields detection (YOLO-based) (commonforms.simplepdf.com)

CommonForms, a new browser-based tool from Joe Barrow, automatically detects PDF form fields (text boxes, checkboxes, signature areas) using a YOLO-style object detection model. The project trains FFDNet-derived models on a large dataset of form layouts, exports them to ONNX, and runs inference entirely client-side with ONNX Runtime Web. That means field detection happens in the browser (WASM/WebGPU backends), avoiding server uploads and enabling instant, privacy-preserving annotation and extraction. For the AI/ML community this demonstrates a practical, low-latency deployment pattern: train a detection model on diverse form data, export to ONNX, and run in-browser for real-time UX and privacy gains. Technically, the approach outputs bounding boxes and class labels for common widget types, which can be mapped back to PDF coordinates for autofill, accessibility tooling, or downstream OCR. Trade-offs include model performance tied to training-set variety (layout styles, scanned vs native PDFs, rotated elements) and typical object-detection failure modes (false positives/negatives). Still, CommonForms is a useful reference for building client-side ML form pipelines and shows how ONNX Runtime Web can deliver production-grade detection without server infrastructure.