Holi-Spatial: Evolving Video Streams into Holistic 3D Spatial Intelligence

Dimensions overall score 8.0

GitHub Code Pulse

No public code linked for this paper yet.

• Holi-Spatial is the first fully automated, large-scale, spatially-aware multimodal dataset constructed from raw video inputs without human intervention.method 95%
  Evidencepartial

  In this work, we propose Holi-Spatial, the first fully automated, large-scale, spatially-aware multimodal dataset, constructed from raw video inputs without human intervention, using the proposed data curation pipeline.

  Implicationpartial

  Explicitly stated in the abstract as a primary contribution of the work.

  Verificationpartial

  partial

• Existing approaches for spatial intelligence benchmarks are severely constrained in scalability due to reliance on manually annotated datasets.limitation 85%
  Evidencepartial

  As a result, their scalability is severely constrained, and model performance is further hindered by domain gaps inherent in these narrowly curated datasets.

  Implicationpartial

  Directly stated in the abstract as a limitation of prior work, though specific comparative scalability metrics are not provided.

  Verificationpartial

  partial

• The Holi-Spatial dataset supports multi-level spatial supervision including 3D Gaussian Splatting reconstructions, object-level and relational semantic annotations, and spatial QA pairs.technical 90%
  Evidencepartial

  Holi-Spatial supports multi-level spatial supervision, ranging from geometrically accurate 3D Gaussian Splatting (3DGS) reconstructions with rendered depth maps to object-level and relational semantic annotations, together with corresponding spatial Question-Answer (QA) pairs.

  Implicationpartial

  Explicitly listed in the abstract as key components of the dataset.

  Verificationpartial

  partial

• The constructed Holi-Spatial-4M dataset contains 12K optimized 3DGS scenes, 1.3M 2D masks, 320K 3D bounding boxes, 320K instance captions, 1.2M 3D grounding instances, and 1.2M spatial QA pairs.result 95%
  Evidencepartial

  we further construct Holi-Spatial-4M, the first large-scale, high-quality 3D semantic dataset, containing 12K optimized 3DGS scenes, 1.3M 2D masks, 320K 3D bounding boxes, 320K instance captions, 1.2M 3D grounding instances, and 1.2M spatial QA pairs

  Implicationpartial

  Specific quantitative details are provided in the abstract, making this a clear, verifiable claim.

  Verificationpartial

  partial

• Holi-Spatial demonstrates exceptional performance in data curation quality, significantly outperforming existing feed-forward and per-scene optimized methods on datasets such as ScanNet, ScanNet++, and DL3DV.result 85%
  Evidencepartial

  Holi-Spatial demonstrates exceptional performance in data curation quality, significantly outperforming existing feed-forward and per-scene optimized methods on datasets such as ScanNet, ScanNet++, and DL3DV.

  Implicationpartial

  Directly stated in the abstract as a result, though specific performance metrics are not provided in the given text.

  Verificationpartial

  partial

• Fine-tuning Vision-Language Models (VLMs) on spatial reasoning tasks using the Holi-Spatial dataset leads to substantial improvements in model performance.result 80%
  Evidencepartial

  Furthermore, fine-tuning Vision-Language Models (VLMs) on spatial reasoning tasks using this dataset has also led to substantial improvements in model performance.

  Implicationpartial

  Stated as a result in the abstract, but the degree of improvement is not quantified in the provided text.

  Verificationpartial

  partial

• Holi-Spatial-4M is the first large-scale, high-quality 3D semantic dataset.technical 90%
  Evidencepartial

  we further construct Holi-Spatial-4M, the first large-scale, high-quality 3D semantic dataset

  Implicationpartial

  Explicitly claimed as a 'first' in the abstract, supported by the description of its novel automated construction.

  Verificationpartial

  partial

Startup potential card