SurfDist paper published in WACV
- 14 hours ago
- 1 min read
Jackson Borchardt's paper "SurfDist: Interpretable Three-Dimensional Instance Segmentation Using Curved Surface Patches" was published in WACV 2026. It's an approach for segmentation of smooth blobs which are common entities in microscopic biomedical imaging. The trick is to concatenate a vector of parameters that define a geometric surface primitive to the standard voxel/pixel output of a dense encoder-decoder network (like a 3D U-Net), a la the StarDist, StarDist3D, and SplineDist approach, but now with curved surfaces in 3D instead of faceted polytopes.
Jackson and I attended the WACV conference in Tucson to present the work and learn about the latest progress and trends in computer vision. We learned a lot: Video understanding and multimodal understanding are the trendy problems. Yes, transformers still rule the day. Human-defined world structure (aka computer graphics) seems to be creeping back into models, despite the ever-present whisperings of the Bitter Lesson. But I think we'll see more of this approach (like, ahem, this paper). Gaussian splatting, which seems to me to be a more of a computer graphics trick rather than a computer vision idea, seems to be fueling a lot of work, probably because it gets you to good-looking demos really fast. Good demos are highly motivating.
here's the journal link for the paper: https://www.computer.org/csdl/proceedings-article/wacv/2026/551100f541/2ggOhZWP5Go
and here's the paper, non-paywalled:
