OctreeNCA: Single-Pass 184 MP Segmentation on Consumer Hardware

Nick Lemke, John Kalkhof Niklas Babendererde Anirban Mukhopadhyay · 2025

Read OctreeNCA: Single-Pass 184 MP Segmentation on Consumer Hardware by Nick Lemke, John Kalkhof Niklas Babendererde Anirban Mukhopadhyay (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2508-06993.txt.

Description

Nick Lemke and co-authors present OctreeNCA for segmenting very large medical images in this 2025 paper. Scans such as prostate MRI and pathology slides should ideally be read whole. Patch by patch reading loses global context and slows work. Standard networks also consume huge graphics memory on big inputs.

Their model grows from neural cellular automata, which are light and size invariant by design. Local communication alone lacks global knowledge, though. An octree neighborhood lets each cell reach distant context efficiently. A custom CUDA routine cuts memory use further and speeds inference. The system segments 184 megapixel inputs in a single pass on consumer hardware. Memory use fell by about ninety percent against standard designs.

Catalog record

CREATORNick Lemke, John Kalkhof Niklas Babendererde Anirban Mukhopadhyay
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE53 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2508-06993

Browse the full library · About the Tower of Babel