CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs

• URL: http://arxiv.org/abs/2609.08345v1
• Date: Tue, 08 Sep 2026 07:16:37 GMT
• Title: CoVeR: Coverage-Based Token Pruning for Multi-View 3D Reasoning in VLMs
• Abstract summary: 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.
• Score: 58.22241532542095
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 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. However, it produces thousands of redundant visual tokens whose cost grows with every view. Existing visual token pruners fall into two families, each limited in the 3D multi-view setting. Learned importance methods rank tokens by attention or encoder features; because redundancy here is fundamentally spatial, they keep near-duplicate tokens from a few prominent regions and leave most of the scene unrepresented. Voxelization methods improve spatial coverage but cannot enforce an exact token budget and saturate as multi-view observations overlap in 3D, capping retention well below the target. 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. CoVeR selects tokens that collectively cover every region of the scene, and solves the limitations of both families: it enforces an exact per-scene budget, breaks the voxelization saturation plateau, and avoids the near-duplicate selections of learned importance. Extensive experiments show CoVeR outperforms prior SOTAs on all three 3D reasoning benchmarks and generalizes as a plug-and-play module tested across four VLMs. Notably, with only $\approx$8% of visual tokens, it preserves 93.5% of full-token performance, surpassing SOTA by 3.9 percentage points on average across benchmarks.

Related papers

• CoverPrune: Coverage-Driven Token Pruning for 3D VLMs via Optimal Transport [50.2610873651284]
  Existing token pruning methods rely on diversity-based selection, discarding similar tokens to maximize dispersion.<n>In 3D environments, this approach frequently drops representative prototype tokens in favor of outliers.<n>We propose a paradigm shift for 3D VLM token pruning: from maximizing OT to preserving visual evidence coverage.
  arXiv  Detail & Related papers  (2026-08-13T13:29:15Z)
• CARVE: Cross-Slice Anisotropic Reallocation of Visual Evidence for Efficient 3D Medical Volume Understanding [8.918770994951872]
  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.
  arXiv  Detail & Related papers  (2026-08-05T06:49:49Z)
• Zero-Shot 3D Question Answering via Hierarchical View-to-Token Transportation [9.296275675671636]
  We propose a hierarchical approach for input context collection at both the view and token levels.<n>Specifically, we combine pixel features with camera parameters and assess view importance based on both semantic content and geometric position.<n>We evaluate our framework on three widely used benchmarks, demonstrating significant improvements over existing tuning-free methods and performance comparable to training-based approaches.
  arXiv  Detail & Related papers  (2026-06-02T03:38:51Z)
• SeGPruner: Semantic-Geometric Visual Token Pruner for 3D Question Answering [18.541639027941198]
  SeGPruner is a semantic-aware and geometry-guided token reduction framework for efficient 3D with multi-view images.<n>It substantially improves inference efficiency, reducing the visual token budget by 91% and inference latency by 86%.
  arXiv  Detail & Related papers  (2026-03-31T08:44:04Z)
• DriveTok: 3D Driving Scene Tokenization for Unified Multi-View Reconstruction and Understanding [75.41918200553525]
  DriveTok is an efficient 3D driving scene tokenizer for unified multi-view reconstruction and understanding.<n>For decoding, we employ a multi-view transformer to reconstruct multi-view features from the scene tokens.<n>We also add a 3D head directly on the scene tokens for 3D semantic occupancy prediction.
  arXiv  Detail & Related papers  (2026-03-19T17:58:22Z)
• Scenes as Tokens: Multi-Scale Normal Distributions Transform Tokenizer for General 3D Vision-Language Understanding [39.18256367776712]
  We present NDTokenizer3D, a 3D VLM that performs a wide range of 3D scene understanding tasks while naturally supporting human interactions.<n>The core of our approach is a novel three-stage scene tokenization pipeline built upon a Multi-Scale Normal Distributions Transform (NDT) representation.<n>NDTokenizer3D first constructs a multi-scale NDT representation from raw high-resolution point clouds, preserving both global context and fine-grained geometric details.
  arXiv  Detail & Related papers  (2025-11-26T09:12:17Z)
• Self-Supervised and Generalizable Tokenization for CLIP-Based 3D Understanding [87.68271178167373]
  We present a universal 3D tokenizer designed for scale-invariant representation learning with a frozen CLIP backbone.<n>S4Token is a tokenization pipeline that produces semantically-informed tokens regardless of scene scale.
  arXiv  Detail & Related papers  (2025-05-24T18:26:30Z)
• RoScenes: A Large-scale Multi-view 3D Dataset for Roadside Perception [98.76525636842177]
  RoScenes is the largest multi-view roadside perception dataset. Our dataset achieves surprising 21.13M 3D annotations within 64,000 $m2$.
  arXiv  Detail & Related papers  (2024-05-16T08:06:52Z)
• Self-supervised Video Representation Learning by Uncovering Spatio-temporal Statistics [74.6968179473212]
  This paper proposes a novel pretext task to address the self-supervised learning problem. We compute a series of partitioning-temporal statistical summaries, such as the spatial location and dominant direction of the largest motion. A neural network is built and trained to yield the statistical summaries given the video frames as inputs.
  arXiv  Detail & Related papers  (2020-08-31T08:31:56Z)

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.