Seeing Once is Enough? Online Geometry-Aware Token Pruning for 3D Question Answering

• URL: http://arxiv.org/abs/2607.04079v1
• Date: Sun, 05 Jul 2026 02:00:54 GMT
• Title: Seeing Once is Enough? Online Geometry-Aware Token Pruning for 3D Question Answering
• Abstract summary: We propose the first online token-pruning method that can be integrated seamlessly with current MLLM models for 3D question answering tasks.<n>Our method enables efficient online processing while reducing up to 50% of token usage.<n>We apply this approach to Qwen2.5-VL-7B and Qwen3-VL-8B, demonstrating improved performance on the ScanQA, SQA3D, and OpenEQA-HM3D benchmarks.
• Score: 26.954138767457277
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Recent Multi-modal Large Language Models (MLLMs) have demonstrated remarkable performance on 2D question answering tasks. However, extending these models to the 3D question answering remains challenging, as they typically require multiple views of the scene, which incurs substantial computational cost at inference. To mitigate this issue, existing solutions rely on strategic frame selection or token-merging algorithms that require preprocessing in advance all frames of the scene, i.e., an offline fashion. In contrast, we propose the first online token-pruning method that can be integrated seamlessly with current MLLM models for 3D question answering tasks, without additional training and with lower memory usage.Our key insight is to project each input frame into a shared voxel space using depth information and camera pose, identifying spatially-overlapped regions across frames and selectively pruning redundant image tokens before they enter the language model. Our method enables efficient online processing while reducing up to 50% of token usage. We apply this approach to Qwen2.5-VL-7B and Qwen3-VL-8B, demonstrating improved performance on the ScanQA, SQA3D, and OpenEQA-HM3D benchmarks.

Related papers

• Stream3D-VLM: Online 3D Spatial Understanding with Incremental Geometry Priors [54.51965288351254]
  We present an online 3D vision-language model that enables real-time spatial understanding from streaming video.<n>Our approach adopts an autoregressive streaming control modeling based on the LLM's next-token prediction to learn when to respond.<n>To address the scarcity of streaming 3D-language data, we further develop a scalable data generation pipeline.
  arXiv  Detail & Related papers  (2026-06-05T04:16:24Z)
• 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)
• Advancing 3D Scene Understanding with MV-ScanQA Multi-View Reasoning Evaluation and TripAlign Pre-training Dataset [56.533371387182065]
  MV-ScanQA is a novel 3D question answering dataset where 68% of questions explicitly require integrating information from multiple views.<n>We present TripAlign, a large-scale and low-cost 2D-3D-language pre-training corpus containing 1M 2D view, set of 3D objects, text> triplets.<n>We further develop LEGO, a baseline method for the multi-view reasoning challenge in MV-ScanQA, transferring knowledge from pre-trained 2D LVLMs to 3D domain with TripAlign.
  arXiv  Detail & Related papers  (2025-08-14T20:35:59Z)
• 3D Question Answering via only 2D Vision-Language Models [87.41421075243103]
  Large vision-language models (LVLMs) have advanced numerous fields.<n>We explore how to harness their potential to address 3D scene understanding tasks, using 3D question answering (3D-QA) as a representative example.<n>Specifically, we sample 2D views from a 3D point cloud and feed them into 2D models to answer a given question.<n>We propose cdViews, a novel approach to automatically selecting critical and diverse Views for 3D-QA.
  arXiv  Detail & Related papers  (2025-05-28T09:04:39Z)
• EmbodiedSAM: Online Segment Any 3D Thing in Real Time [61.2321497708998]
  Embodied tasks require the agent to fully understand 3D scenes simultaneously with its exploration.<n>An online, real-time, fine-grained and highly-generalized 3D perception model is desperately needed.
  arXiv  Detail & Related papers  (2024-08-21T17:57:06Z)
• Coarse Correspondences Boost Spatial-Temporal Reasoning in Multimodal Language Model [51.83436609094658]
  We introduce Coarse Correspondences, a simple lightweight method that enhances MLLMs' spatial-temporal reasoning with 2D images as input. Our method uses a lightweight tracking model to identify primary object correspondences between frames in a video or across different image viewpoints. We demonstrate that this simple training-free approach brings substantial gains to GPT4-V/O consistently on four benchmarks.
  arXiv  Detail & Related papers  (2024-08-01T17:57:12Z)
• Bridging the Gap between 2D and 3D Visual Question Answering: A Fusion Approach for 3D VQA [6.697298321551588]
  In 3D Visual Question Answering (3D VQA), the scarcity of fully annotated data and limited visual content diversity hampers the generalization to novel scenes and 3D concepts. We propose question-conditional 2D view selection procedure, pinpointing semantically relevant 2D inputs for crucial visual clues. We then integrate this 2D knowledge into the 3D-VQA system via a two-branch Transformer structure.
  arXiv  Detail & Related papers  (2024-02-24T23:31:34Z)
• EgoLoc: Revisiting 3D Object Localization from Egocentric Videos with Visual Queries [68.75400888770793]
  We formalize a pipeline that better entangles 3D multiview geometry with 2D object retrieval from egocentric videos. Specifically, our approach achieves an overall success rate of up to 87.12%, which sets a new state-of-the-art result in the VQ3D task.
  arXiv  Detail & Related papers  (2022-12-14T01:28:12Z)

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.