Research map / 3D vision / 3D generation
DINOv3D: 2D-3D Joint Optimization for Unified Spatial Understanding
Bo Zhou, Jianzhe Gao, Zhihui Wang, Lingxiang Wu, Jinqiao Wang, Yazhou Yao, Wenguan Wang
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 dual-branch teacher-student architecture to frozen baseline (ablation) | Frozen baseline · ScanNet | PSNR: 23.45 -> 24.62 | Improved |
| add dual-branch architecture (ablation) | Frozen baseline · ADE20k | mIoU: 54.9 -> 55.0 | No clear effect |
| add ray-depth-semantic alignment (ablation) | Dual-branch model · ScanNet | PSNR: 24.62 -> 25.41 | Improved |
| add ray-depth-semantic alignment (ablation) | Dual-branch model · ScanNet | OVSS mIoU: 54.5 -> 55.6 | Improved |
| add long context (128 frames) (ablation) | Dual-branch + alignment · ScanNet | PSNR: 25.41 -> 25.58 | Improved |
| train on Uni3D dataset (ablation) | Full pipeline · ScanNet | mIoU: 56.5 -> 57.7 | Improved |
| extend context length from 8 to 128 frames (ablation) | DINOv3D · ScanNet | PSNR: 24.62 -> 25.67 | Improved |
| extend context length from 128 to 144 frames (ablation) | DINOv3D · ScanNet | PSNR: 25.67 -> 25.69 | No clear effect |
| use LoRA instead of frozen backbone (ablation) | DINOv3D · ScanNet | PSNR: 24.15 -> 25.67 | Improved |
| use LoRA instead of full fine-tuning (ablation) | DINOv3D · NYUv2 | RMSE: 0.385 -> 0.324 | Improved |
| use LoRA instead of full fine-tuning (ablation) | DINOv3D · ScanNet | PSNR: 25.72 -> 25.67 | No clear effect |
| use LoRA instead of full fine-tuning (ablation) | DINOv3D · NYUv2 | mAcc: 77.5 -> 84.6 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Uni3R — Uni3R: Unified 3D Reconstruction and Semantic Understanding via Generalizable Gaussian Splatting from Unposed Multi-View Images (CVPR 2026) · better on 6 of 6 shared results
- MVSplat — MVSplat: Efficient 3D Gaussian Splatting from Sparse Multi-View Images (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Imagine3D-LLM: Teaching MLLMs to Imagine 3D Scenes Before Answering (NeurIPS 2026 · similarity 0.90)
- Beyond 3D VQAs: Injecting 3D Spatial Priors into Vision-Language Models for Enhanced Geometric Reasoning (CVPR 2026 · similarity 0.89)
- Cross-Modal and Uncertainty-Aware Agglomeration for Open-Vocabulary 3D Scene Understanding (CVPR 2025 · similarity 0.89)
- G$^2$VLM: Geometry Grounded Vision Language Model with Unified 3D Reconstruction and Spatial Reasoning (CVPR 2026 · similarity 0.89)
- From Flatland to Space: Teaching Vision-Language Models to Perceive and Reason in 3D (NeurIPS 2025 · similarity 0.89)
- Uncovering and Shaping the Latent Representation of 3D Scene Topology in Vision-Language Models (NeurIPS 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.
