Research map / Vision-language and multimodal / Multimodal LLMs
3D-Aware VLMs with Implicit and Explicit Geometries
Wenhao Li, Xueying Jiang, Quanhao Qian, Deli Zhao, Ran Xu, Shijian Lu, Gongjie Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add implicit geometry tokens (IGTs) (ablation) | Qwen2.5-VL-3B · 3D video detection | F1_25: 30.9 -> 40.5 | Improved |
| add explicit geometry tokens (EGTs) (ablation) | Qwen2.5-VL-3B · 3D video detection | F1_25: 30.9 -> 34.7 | Improved |
| add EGTs on top of IGTs (ablation) | VLM-IE3D · 3D video detection | F1_25: 40.5 -> 42.8 | Improved |
| fuse IGTs and EGTs with implicit-explicit attention instead of concat (ablation) | VLM-IE3D · 3D video detection | F1_25: 41.5 -> 42.8 | Improved |
| fuse with IEA instead of simple addition (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.4 -> 42.8 | Improved |
| use learnable weighted fusion instead of IEA (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.8 -> 41.2 | Hurt |
| use depth, point or Gaussian attribute for EGTs (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.5 to 42.8 | No clear effect |
| use deep encoder (DepthAnything V2) for explicit embedding instead of light patch embedding (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.8 -> 35.9 | Hurt |
| use average pooling instead of one-layer patch embedding for EGTs (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.8 -> 42.0 | Hurt |
| swap AnySplat for VGGT or pi^3 as geometry encoder (ablation) | VLM-IE3D · 3D video detection | F1_25: 42.8 -> 41.7 (VGGT), 42.1 (pi^3) | Mixed |
| add implicit and explicit geometry to VLM | Qwen2.5-VL-3B · Scan2Cap | C@0.5: 58.0 -> 80.4 | Improved |
| add EGTs on top of implicit-only VG LLM | VG LLM · ScanRefer Acc@0.50 | Acc@0.50: 11.8 -> 16.9 | Improved |
Closest papers by meaning
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.91)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.91)
- VLM-3R: Vision-Language Models Augmented with Instruction-Aligned 3D Reconstruction (CVPR 2026 · similarity 0.90)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.90)
- HiSpatial: Taming Hierarchical 3D Spatial Understanding in Vision-Language Models (CVPR 2026 · similarity 0.90)
- Abstract 3D Perception for Spatial Intelligence in Vision-Language Models (CVPR 2026 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
