--- title: Polyp Segmentation in the Browser emoji: 🔬 colorFrom: pink colorTo: indigo sdk: static app_file: index.html pinned: false license: mit short_description: Kvasir-SEG polyp U-Net with uncertainty, in your browser models: - haroonx02/polyp-segmentation-unet-resnet34 custom_headers: cross-origin-embedder-policy: credentialless cross-origin-opener-policy: same-origin cross-origin-resource-policy: cross-origin --- A U-Net polyp segmentation model (ResNet-34 encoder, trained on Kvasir-SEG) exported to ONNX, quantised to INT8 and run client-side with onnxruntime-web (WebAssembly). A second graph keeps decoder dropout on and returns a Monte-Carlo uncertainty map. Pre- and post-processing are ported to JavaScript and checked against the Python pipeline. Research demo only, not for clinical use. Code, training, evaluation and error analysis: [github.com/haroonkhan02/polyp-segmentation-unet](https://github.com/haroonkhan02/polyp-segmentation-unet).