CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding

• URL: http://arxiv.org/abs/2608.04515v1
• Date: Wed, 05 Aug 2026 06:49:49 GMT
• Title: CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding
• Abstract summary: Slice-based MLLMs leverage mature 2D encoders by representing 3D volumes as sequences of 2D slices.<n>This slice-wise formulation produces thousands of visual tokens that burden the LLM backbone.<n>We present CARVE, a training-free framework that compresses visual tokens prior to LLM inference and casts token reduction as budget-constrained 2.5D allocation.
• Score: 8.918770994951872
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Slice-based MLLMs leverage mature 2D encoders by representing 3D volumes as sequences of 2D slices. However, this slice-wise formulation produces thousands of visual tokens that burden the LLM backbone, many of which capture overlapping visual evidence across adjacent slices. To understand how effectively a growing visual token budget improves performance, we perform scaling analyses on two 3D medical VQA benchmarks and find diminishing returns: cost keeps rising while accuracy saturates, and improving in-plane resolution is more effective than adding slices at comparable budgets. The budget should therefore be allocated more selectively rather than simply enlarged, yet most token compression methods are designed for 2D images or videos, where redundancy arises from spatial layout or temporal motion rather than from near-duplicate content along the depth axis. We present CARVE, a training-free framework that compresses visual tokens prior to LLM inference and casts token reduction as budget-constrained 2.5D allocation. CARVE partitions the depth axis into coherent windows and allocates tokens non-uniformly according to normalized cross-slice evidence. Under a shared budget, CARVE builds spatial anchors on representative slices and retrieves locally varying evidence from the full volume, then merges remaining eligible tokens into nearby anchors within each window. Removing roughly 80% of the visual tokens on Hulu-Med-7B, CARVE leads all compression baselines on every AMOS-MM report-generation metric, with 6.2 points higher retention of full-token quality than the strongest baseline, and preserves 98.1% of full-token performance across three VQA benchmarks.

Related papers

• CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs [58.22241532542095]
  Representing a 3D scene as multi-view images allows 2D VLMs to reason in 3D by reusing priors from pre-training, sidestepping the scarcity of annotated 3D data.<n>Existing visual token pruners fall into two families, each limited in the 3D multi-view setting.<n>We show that spatial coverage is associated with 3D reasoning performance and introduce CoVeR, a deterministic, training-free selector that uses only token coordinates, with no learned signals.
  arXiv  Detail & Related papers  (2026-09-08T07:16:37Z)
• HiSC: Hierarchical Spatial Clustering Token Compression for Efficient 3D Scene Understanding [16.080049512272026]
  3D vision-language models (3D VLMs) enable spatial reasoning over multi-view scenes but suffer from substantial token redundancy.<n>We propose textbfHiSC, a training-free framework for hierarchical spatial clustering token compression in 3D VLMs.<n>HiSC lifts token compression from token-level selection to cluster-level processing by organizing tokens into spatially grounded clusters.
  arXiv  Detail & Related papers  (2026-08-05T09:15:43Z)
• MedARC: Training-Free Adaptive Redundancy Compression of Visual Tokens for 3D Medical Vision-Language Models [3.8740159612186265]
  We propose MedARC, a framework for Adaptive Redundancy Compression of visual tokens in 3D medical vision-language models.<n>MedARC estimates token importance by integrating three complementary cues.<n>Experiments on CT-RATE and MR-RATE show that MedARC reduces visual-token overhead and inference time while preserving or improving diagnostic performance.
  arXiv  Detail & Related papers  (2026-07-29T07:27:09Z)
• Efficient3D: A Unified Framework for Adaptive and Debiased Token Reduction in 3D MLLMs [37.792545310147965]
  This paper presents Efficient3D, a unified framework for visual token pruning.<n>It achieves superior performance compared with unpruned baselines, with a +2.57% CIDEr improvement on the Scan2Cap dataset.
  arXiv  Detail & Related papers  (2026-04-03T03:32:55Z)
• MedPruner: Training-Free Hierarchical Token Pruning for Efficient 3D Medical Image Understanding in Vision-Language Models [59.180043227905294]
  We propose MedPruner, a training-free and model-agnostic hierarchical token pruning framework for efficient 3D medical image understanding.<n>We show that MedPruner enables models such as MedGemma to maintain or even exceed their original performance while retaining fewer than 5% of visual tokens.
  arXiv  Detail & Related papers  (2026-03-12T07:37:00Z)
• Prune Redundancy, Preserve Essence: Vision Token Compression in VLMs via Synergistic Importance-Diversity [54.95089105944234]
  Vision-language models (VLMs) face significant computational inefficiencies caused by excessive generation of visual tokens.<n>We propose PruneSID, a training-free Synergistic Importance-Diversity approach featuring a two-stage pipeline.<n>We demonstrate state-of-the-art performance, achieving 96.3% accuracy on LLaVA-1.5 with only 11.1% token retention, and 92.8% accuracy at extreme compression rates (5.6%) on LLaVA-NeXT.
  arXiv  Detail & Related papers  (2026-03-10T10:31:58Z)
• Nüwa: Mending the Spatial Integrity Torn by VLM Token Pruning [82.39668822222386]
  Vision token pruning has proven to be an effective acceleration technique for the efficient Vision Language Model (VLM)<n>We propose $textNwa$, a two-stage token pruning framework that enables efficient feature aggregation while maintaining spatial integrity.<n>Experiments demonstrate that $textNwa$ achieves SOTA performance on multiple VQA benchmarks (from 94% to 95%) and yields substantial improvements on visual grounding tasks (from 7% to 47%)
  arXiv  Detail & Related papers  (2026-02-03T00:51:03Z)
• Multimodal Visual Surrogate Compression for Alzheimer's Disease Classification [69.87877580725768]
  Multimodal Visual Surrogate Compression (MVSC) learns to compress and adapt large 3D sMRI volumes into compact 2D features.<n>MVSC has two key components: a Volume Context that captures global cross-slice context under textual guidance, and an Adaptive Slice Fusion module that aggregates slice-level information in a text-enhanced, patch-wise manner.
  arXiv  Detail & Related papers  (2026-01-29T13:05:46Z)
• HCC-3D: Hierarchical Compensatory Compression for 98% 3D Token Reduction in Vision-Language Models [19.63602285036466]
  3D understanding has drawn significant attention recently, leveraging Vision-Language Models (VLMs) to enable multi-modal reasoning between point cloud and text data.<n>Current 3D-VLMs directly embed the 3D point clouds into 3D tokens, following large 2D-VLMs with powerful reasoning capabilities.<n>This framework has a great computational cost limiting its application, where we identify that the bottleneck lies in processing all 3D tokens in the Large Language Model (LLM) part.<n>This raises the question: how can we reduce the computational overhead introduced by 3D tokens while preserving the integrity of their essential information?<n>
  arXiv  Detail & Related papers  (2025-11-13T02:28:10Z)
• econSG: Efficient and Multi-view Consistent Open-Vocabulary 3D Semantic Gaussians [56.85804719947]
  We propose econSG for open-vocabulary semantic segmentation with 3DGS.<n>Our econSG shows state-of-the-art performance on four benchmark datasets compared to the existing methods.
  arXiv  Detail & Related papers  (2025-04-08T13:12:31Z)
• UNETR++: Delving into Efficient and Accurate 3D Medical Image Segmentation [93.88170217725805]
  We propose a 3D medical image segmentation approach, named UNETR++, that offers both high-quality segmentation masks as well as efficiency in terms of parameters, compute cost, and inference speed. The core of our design is the introduction of a novel efficient paired attention (EPA) block that efficiently learns spatial and channel-wise discriminative features. Our evaluations on five benchmarks, Synapse, BTCV, ACDC, BRaTs, and Decathlon-Lung, reveal the effectiveness of our contributions in terms of both efficiency and accuracy.
  arXiv  Detail & Related papers  (2022-12-08T18:59:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.