HCC-3D: Hierarchical Compensatory Compression for 98% 3D Token Reduction in Vision-Language Models

• URL: http://arxiv.org/abs/2511.09883v1
• Date: Fri, 14 Nov 2025 01:15:39 GMT
• Title: HCC-3D: Hierarchical Compensatory Compression for 98% 3D Token Reduction in Vision-Language Models
• Abstract summary: 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>
• Score: 19.63602285036466
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: 3D understanding has drawn significant attention recently, leveraging Vision-Language Models (VLMs) to enable multi-modal reasoning between point cloud and text data. Current 3D-VLMs directly embed the 3D point clouds into 3D tokens, following large 2D-VLMs with powerful reasoning capabilities. However, 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. This raises the question: how can we reduce the computational overhead introduced by 3D tokens while preserving the integrity of their essential information? To address this question, we introduce Hierarchical Compensatory Compression (HCC-3D) to efficiently compress 3D tokens while maintaining critical detail retention. Specifically, we first propose a global structure compression (GSC), in which we design global queries to compress all 3D tokens into a few key tokens while keeping overall structural information. Then, to compensate for the information loss in GSC, we further propose an adaptive detail mining (ADM) module that selectively recompresses salient but under-attended features through complementary scoring. Extensive experiments demonstrate that HCC-3D not only achieves extreme compression ratios (approximately 98%) compared to previous 3D-VLMs, but also achieves new state-of-the-art performance, showing the great improvements on both efficiency and performance.

Related papers

• 3DZip: Spatial-Aware Feature Diversity-Guided Token Compression for 3D Question Answering [8.79576142391319]
  Recent 3D vision-language models construct geometry aware tokens by projecting 2D visual features into world coordinates.<n>This design generates thousands of tokens per scene, resulting in substantial computational and memory overhead.<n>We propose 3DZip, a three-stage token compression framework that first applies coarse voxelization to remove point-level redundancy, then selects anchor tokens based on feature-space diversity, and finally merges remaining tokens under spatial constraints to preserve geometric coherence.
  arXiv  Detail & Related papers  (2026-08-02T12:11:51Z)
• Lift3D-VLA: Lifting VLA Models to 3D Geometry and Dynamics-Aware Manipulation [76.16065370700488]
  We introduce Lift3D-VLA, a unified VLA framework that equips models with explicit 3D point cloud reasoning and enables temporally coherent action generation.<n>We show that Lift3D-VLA achieves 10.8% and 11.1% higher mean success rates on MetaWorld and RLBench than the best-performing prior VLA methods.
  arXiv  Detail & Related papers  (2026-07-07T17:59:47Z)
• 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)
• PatchAlign3D: Local Feature Alignment for Dense 3D Shape understanding [67.15800065888887]
  Current foundation models for 3D shapes excel at global tasks (retrieval, classification) but transfer poorly to local part-level reasoning.<n>We introduce an encoder-only 3D model that produces language-aligned patch-level features directly from point clouds.<n>Our 3D encoder achieves zero-shot 3D part segmentation with fast single-pass inference without any test-time multi-view rendering.
  arXiv  Detail & Related papers  (2026-01-05T18:55:45Z)
• C3G: Learning Compact 3D Representations with 2K Gaussians [55.04010158339562]
  Recent approaches use per-pixel 3D Gaussian Splatting for reconstruction, followed by a 2D-to-3D feature lifting stage for scene understanding.<n>We propose C3G, a novel feed-forward framework that estimates compact 3D Gaussians only at essential spatial locations.
  arXiv  Detail & Related papers  (2025-12-03T17:59:05Z)
• 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)
• 3D-R1: Enhancing Reasoning in 3D VLMs for Unified Scene Understanding [11.069512983766783]
  Large vision-language models (VLMs) have made significant strides in 2D visual understanding tasks.<n>We propose 3D-R1, a foundation model that enhances the reasoning capabilities of 3D VLMs.<n>Extensive experiments demonstrate that 3D-R1 delivers an average improvement of 10% across various 3D scene benchmarks.
  arXiv  Detail & Related papers  (2025-07-31T11:59:06Z)
• Fast3D: Accelerating 3D Multi-modal Large Language Models for Efficient 3D Scene Understanding [24.964149224068027]
  We propose Fast3D, a plug-and-play visual token pruning framework for 3D MLLMs.<n>Global Attention Prediction (GAP) learns to predict the global attention distributions of the target model, enabling efficient token importance estimation.<n>SAP, introduces dynamic token budgets through attention-based complexity assessment, automatically adjusting layer-wise pruning ratios.
  arXiv  Detail & Related papers  (2025-07-12T16:29:02Z)
• Does Your 3D Encoder Really Work? When Pretrain-SFT from 2D VLMs Meets 3D VLMs [72.11701578308804]
  This paper categorizes recent 3D Vision-Language Models into 3D object-centric, 2D image-based, and 3D scene-centric approaches.<n>Despite the architectural similarity of 3D scene-centric VLMs to their 2D counterparts, they have exhibited comparatively lower performance compared with the latest 3D object-centric and 2D image-based approaches.<n>Our investigation suggests that while these models possess cross-modal alignment capabilities, they tend to over-rely on linguistic cues and overfit to frequent answer distributions.
  arXiv  Detail & Related papers  (2025-06-05T17:56:12Z)
• AdaToken-3D: Dynamic Spatial Gating for Efficient 3D Large Multimodal-Models Reasoning [27.40106634796608]
  Large Multimodal Models (LMMs) have become a pivotal research focus in deep learning.<n>Currently, 3D LMMs employing thousands of spatial tokens for multimodal reasoning suffer from critical inefficiencies.<n>We propose AdaToken-3D, an adaptive spatial token optimization framework that dynamically prunes redundant tokens.
  arXiv  Detail & Related papers  (2025-05-19T07:11:07Z)
• 3UR-LLM: An End-to-End Multimodal Large Language Model for 3D Scene Understanding [49.15555885075644]
  We develop pipeline based on open-source 2D MLLMs and LLMs to generate high-quality 3D-text pairs.<n>We introduce the 3UR-LLM model, an end-to-end 3D MLLM designed for precise interpretation of 3D scenes.
  arXiv  Detail & Related papers  (2025-01-14T03:50:23Z)
• Make Your ViT-based Multi-view 3D Detectors Faster via Token Compression [78.93023152602525]
  Slow inference speed is one of the most crucial concerns for deploying multi-view 3D detectors to tasks with high real-time requirements like autonomous driving. We propose a simple yet effective method called TokenCompression3D (ToC3D) Our method can nearly maintain the performance of recent SOTA with up to 30% inference speedup, and the improvements are consistent after scaling up the ViT and input resolution.
  arXiv  Detail & Related papers  (2024-09-01T06:58:08Z)
• DatasetNeRF: Efficient 3D-aware Data Factory with Generative Radiance Fields [68.94868475824575]
  This paper introduces a novel approach capable of generating infinite, high-quality 3D-consistent 2D annotations alongside 3D point cloud segmentations. We leverage the strong semantic prior within a 3D generative model to train a semantic decoder. Once trained, the decoder efficiently generalizes across the latent space, enabling the generation of infinite data.
  arXiv  Detail & Related papers  (2023-11-18T21:58:28Z)
• 3D Sketch-aware Semantic Scene Completion via Semi-supervised Structure Prior [50.73148041205675]
  The goal of the Semantic Scene Completion (SSC) task is to simultaneously predict a completed 3D voxel representation of volumetric occupancy and semantic labels of objects in the scene from a single-view observation. We propose to devise a new geometry-based strategy to embed depth information with low-resolution voxel representation. Our proposed geometric embedding works better than the depth feature learning from habitual SSC frameworks.
  arXiv  Detail & Related papers  (2020-03-31T09:33:46Z)

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.