Our paper on elevation-aware flood extent mapping brings topologically guided segmentation to the web for crowdsourced labeling. ML flood mapping models trained on these maps w/ our correction framework get similar quality as those w/ solely expert labels! PDF: www.willusher.io/publications...

Our new paper, Approximate Puzzlepiece Compositing, applies OIT to data-parallel rendering to efficiently render overlapping, jigsaw puzzle-like data distributions, without repartitioning or expensive compositing. It'll presented at Pacific Vis 2025, check it out: www.willusher.io/publications... !

I had kept this as my pinned tweet, it's 5yrs old but still relevant today! My post + interactive tool on the RTX shader binding table to explain it across #DXR / #Vulkan / #OptiX www.willusher.io/graphics/201... . The tool can setup an SBT, scene, and call trace ray to see what shaders are called!

Excited to share some excellent recent work with Landon Dyken and Sidharth Kumar: www.willusher.io/publications..., we extend our progressive raycasting pipeline w/ an ML image completion step to stop raycasting early and complete the image with ML! It still runs in the browser using ONNX #WebGPU !

Finally got back to my "0 to #glTF with #WebGPU" series, this post looks at adding basic material and texture support to get nicer looking renderings (and more interesting workloads to try and batch up later): www.willusher.io/graphics/202... code: github.com/Twinklebear/...

I just published a blog about implementing marching cubes in #WebGPU and benchmarking it against a native #Vulkan implementation: www.willusher.io/graphics/202..., as a spoiler, the WebGPU version performs on par with the native Vulkan one!