Research map / Vision-language and multimodal / Multimodal LLMs
Towers of Babel: Combining Images, Language, and 3D Geometry for Learning Multimodal Vision
Xiaoshi Wu, Hadar Averbuch-Elor, Jin Sun, Noah Snavely
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 3D contrastive loss (ablation) | classification baseline · WikiScenes WS-K | mAP: 70.8 -> 75.3 | Improved |
| add 3D contrastive loss (ablation) | classification baseline · WikiScenes WS-U | mAP: 48.3 -> 52.0 | Improved |
| use inter-image negative sampling instead of 3D MSE loss (ablation) | 3D consistency loss · WikiScenes WS-K | mAP: 71.4 -> 75.3 | Improved |
| use contrastive loss instead of 3D triplet loss (ablation) | 3D consistency loss · WikiScenes WS-K | mAP: 72.1 -> 75.3 | Improved |
| 3D MSE loss added to baseline (ablation) | classification baseline · WikiScenes WS-U | mAP: 48.3 -> 49.5 | Improved |
| intra-image vs inter-image negative sampling (ablation) | 3D contrastive loss · WikiScenes WS-U | mAP: 52.5 vs 52.0 | No clear effect |
| add pixel-level self-supervised loss (ablation) | semantic classification loss · WikiScenes WS-U | mAP: 53.8 -> 52.0 | No clear effect |
| add pixel-level self-supervised loss (ablation) | semantic classification loss · WikiScenes WS-K | mAP: 73.4 -> 75.3 | Improved |
| add 3D contrastive loss | ResNet-50 · WikiScenes WS-K | mAP: 68.5 -> 73.9 | Improved |
| add 3D contrastive loss | MobileNetV2 · WikiScenes WS-U | mAP: 50.2 -> 53.4 | Improved |
| add 3D contrastive loss | 2D segmentation model · WikiScenes segmentation | precision: 68.6 -> 80.8 | Improved |
| add 3D contrastive loss | 2D segmentation model · WikiScenes segmentation | IoU: 25.4 -> 27.2 | Improved |
Closest papers by meaning
- OpenLex3D: A Tiered Benchmark for Open-Vocabulary 3D Scene Representations (NeurIPS 2025 · similarity 0.87)
- MMScan: A Multi-Modal 3D Scene Dataset with Hierarchical Grounded Language Annotations (NeurIPS 2024 · similarity 0.87)
- Merge3D: Efficient 3D Multimodal LLMs via Joint 2D-3D Token Merging (CVPR 2026 · similarity 0.86)
- Learning from Videos for 3D World: Enhancing MLLMs with 3D Vision Geometry Priors (NeurIPS 2025 · similarity 0.86)
- Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering (NeurIPS 2026 · similarity 0.86)
- MLLM-For3D: Adapting Multimodal Large Language Model for 3D Reasoning Segmentation (NeurIPS 2025 · similarity 0.86)
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.
