Skip to content
Tagged articles

3D Scene Understanding

3 articles · Page 1 of 1
Machine Heart
Machine Heart
Oct 20, 2026 · Artificial Intelligence

VLMs Encode 3D Scene Topology in Latent Space: NeurIPS 2026 Spotlight Findings

University of Pittsburgh researchers show that vision-language models internally represent 3D scene topology in middle-layer latent spaces, revealed by separating appearance attributes and visualizing spatial structure; Dirichlet energy quantifies neighborhood preservation, and fine-tuning with this loss improves spatial reasoning on benchmarks like VSI-Bench.

3D Scene UnderstandingDirichlet EnergyInterpretability
0 likes · 12 min read
VLMs Encode 3D Scene Topology in Latent Space: NeurIPS 2026 Spotlight Findings
Machine Heart
Machine Heart
Apr 30, 2026 · Artificial Intelligence

Can Internet Videos Replace 3D Annotations? Introducing SceneVerse++ – the Largest Real‑World 3D Scene Dataset

The BIGAI team presents SceneVerse++, a massive real‑world indoor 3D scene dataset built from unlabelled internet videos via an automated pipeline, and demonstrates substantial zero‑shot and fine‑tuned performance gains on 3D detection, spatial VQA, and vision‑language navigation tasks.

3D Scene UnderstandingSceneVerse++Vision-Language Navigation
0 likes · 18 min read
Can Internet Videos Replace 3D Annotations? Introducing SceneVerse++ – the Largest Real‑World 3D Scene Dataset
Xiaohongshu Tech REDtech
Xiaohongshu Tech REDtech
May 10, 2024 · Artificial Intelligence

SIF3D: Sense‑Informed Forecasting of 3D Human Motion with Multimodal Attention

SIF3D is a scene‑aware 3D human motion forecasting framework that fuses observed motion, 3D point‑cloud scenes, and gaze through novel ternary intention‑aware and semantic‑coherence‑aware attention mechanisms, encoding with PointNet++ and Transformers, and decoding with a graph‑convolutional network, achieving state‑of‑the‑art results on GIMO and GTA‑1M benchmarks.

3D Scene UnderstandingCVPR2024Computer Vision
0 likes · 15 min read
SIF3D: Sense‑Informed Forecasting of 3D Human Motion with Multimodal Attention