aiqbench analysis
BATS introduces a novel architecture that dynamically allocates computational resources by focusing high-resolution processing on image boundaries while coarsely representing homogeneous regions, significantly reducing memory usage without sacrificing segmentation accuracy. This approach addresses the critical challenge of inference cost in 3D medical imaging models like nnU-Net Revisited, offering a practical path toward more efficient deployment of volumetric AI systems. By decoupling resolution from feature map density based on boundary relevance, the method demonstrates that resource efficiency can be achieved alongside high performance across diverse CT and MRI datasets.
BATS: Resource-Efficient Volumetric Segmentation with Boundary-Aware Mixed-Resolution Tokens
arXiv:2607.26829v1 Announce Type: cross Abstract: Many high-performing volumetric segmentation models maintain dense multi-scale feature maps, leading to high activation memory and inference cost. We present BATS (Boundary-Aware Token Selection), a 3D medical image segmentation architecture that concentrates fine-resolution processing near predicted class boundaries. A dense boundary predictor identifies where additional resolution is needed, while a fine-first context cascade constructs an input-dependent mixed-resolution hierarchy. Homogeneous regions are represented coarsely, with finer tokens retained around boundaries, thin structures, and small targets. The sparse hierarchy is refined and rasterised into a dense segmentation. BATS predicts boundary relevance independently at every resolution level, preventing an erroneous coarse-scale decision from suppressing fine-scale evidence. Parent cluster attention further injects hierarchical ancestor tokens into local attention neighbourhoods, providing cross-scale context without dense multi-scale feature maps or cross-scale neighbour search. We evaluate BATS on five public CT and MRI datasets using the standardised nnU-Net Revisited protocol. BATS achieves the highest LiTS Dice among the compared methods and averages within 0.37 Dice points of the strongest dense baseline, MedNeXt-L, across the five datasets. Relative to MedNeXt-L, it reduces peak allocated GPU memory by more than 53% on KiTS, LiTS, and BraTS. Inference is up to 30% faster on KiTS and LiTS, which retain fewer tokens, but slower on the more token-dense BraTS. Mixed-resolution processing therefore provides consistent memory savings, while runtime and accuracy gains depend on dataset boundary density.
Abstract:Many high-performing volumetric segmentation models maintain dense multi-scale feature maps, leading to high activation memory and inference cost. We present BATS (Boundary-Aware Token Selection), a 3D medical image segmentation architecture that concentrates fine-resolution processing near predicted class boundaries. A dense boundary predictor identifies where additional resolution is needed, while a fine-first context cascade constructs an input-dependent mixed-resolution hierarchy. Homogeneous regions are represented coarsely, with finer tokens retained around boundaries, thin structures, and small targets. The sparse hierarchy is refined and rasterised into a dense segmentation. BATS predicts boundary relevance independently at every resolution level, preventing an erroneous coarse-scale decision from suppressing fine-scale evidence. Parent cluster attention further injects hierarchical ancestor tokens into local attention neighbourhoods, providing cross-scale context without dense multi-scale feature maps or cross-scale neighbour search. We evaluate BATS on five public CT and MRI datasets using the standardised nnU-Net Revisited protocol. BATS achieves the highest LiTS Dice among the compared methods and averages within 0.37 Dice points of the strongest dense baseline, MedNeXt-L, across the five datasets. Relative to MedNeXt-L, it reduces peak allocated GPU memory by more than 53% on KiTS, LiTS, and BraTS. Inference is up to 30% faster on KiTS and LiTS, which retain fewer tokens, but slower on the more token-dense BraTS. Mixed-resolution processing therefore provides consistent memory savings, while runtime and accuracy gains depend on dataset boundary density.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2607.26829 [cs.CV]
(or arXiv:2607.26829v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2607.26829 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 325 words
Related coverage
EA-LiteUNet: An Edge-Adaptive and Resource-Efficient U-Net for Boundary-Sensitive Dermoscopic Image Segmentation
arXiv · Aug 18, 2026
Unlock Exascale Performance on NVIDIA GB200 NVL72 with Slurm Topology-Aware Job Scheduling
NVIDIA · May 21, 2026
DPNeXt: A Lightweight Multi-Scale Feature Fusion Framework for Efficient ViT-Based Multi-Task Dense Prediction
arXiv · Jul 20, 2026
Vertical Advantage: Transforming Industries with Lakebase and Agentic AI
Databricks · Aug 27, 2026