Research map / 3D vision / Depth and stereo
Pixel-Perfect Visual Geometry Estimation
Gangwei Xu, Haotong Lin, Hongcheng Luo, Haiyang Sun, Bing Wang, Guang Chen, Sida Peng, Hangjun Ye, Xin Yang
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 semantics prompt (SP-DiT) to vanilla pixel-space DiT (ablation) | DiT · NYUv2 | AbsRel: 22.5 -> 4.8 | Improved |
| add REPA token alignment to vanilla DiT (ablation) | DiT · NYUv2 | AbsRel: 22.5 -> 17.6 | Improved |
| add Cascade DiT (coarse-to-fine patch size) to SP-DiT (ablation) | SP-DiT · NYUv2 | AbsRel: 4.8 -> 4.3 | Improved |
| add Cascade DiT to SP-DiT (ablation) | SP-DiT · RTX 4090 | inference time: 0.20 -> 0.14 | Improved |
| add Cascade DiT to SP-DiT (ablation) | SP-DiT · ScanNet | AbsRel: 6.2 -> 4.5 | Improved |
| use DINOv2 encoder instead of MAE for semantics (ablation) | SP-DiT · NYUv2 | AbsRel: 6.4 -> 4.8 | Improved |
| use MoGe 2 encoder instead of Depth Anything v2 for semantics (ablation) | SP-DiT · NYUv2 | AbsRel: 4.3 -> 3.3 | Improved |
| add reference-guided token propagation (RGTP) (ablation) | SP-DiT (DAv2) · NYUv2 video | AbsRel: 12.2 -> 7.6 | Improved |
| replace SP-DiT with multi-view SC-DiT (VGGT) with RGTP (ablation) | PPVD · NYUv2 video | AbsRel: 7.6 -> 4.5 | Improved |
| use pi^3 instead of VGGT as multi-view semantic encoder (ablation) | SC-DiT + RGTP · NYUv2 video | AbsRel: 4.5 -> 3.8 | Improved |
| use PPVD instead of Video Depth Anything | video depth · ScanNet | AbsRel: 8.9 -> 3.7 | Improved |
| use pixel-space diffusion instead of VAE latent space (GT(VAE) reconstruction) (ablation) | depth representation · Hypersim edge-aware | Chamfer Distance | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.
- MoGe 2 — MoGe-2: Accurate Monocular Geometry with Metric Scale and Sharp Details (NeurIPS 2025) · better on 2 of 5 shared results
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
- Pixel-Perfect Depth with Semantics-Prompted Diffusion Transformers (NeurIPS 2025 · similarity 0.96)
- Novel View Synthesis with Pixel-Space Diffusion Models (CVPR 2025 · similarity 0.88)
- World Tracing: Pixel-Aligned Geometry Beyond the Visible (NeurIPS 2026 · similarity 0.87)
- SharpDepth: Sharpening Metric Depth Predictions Using Diffusion Distillation (CVPR 2025 · similarity 0.87)
- Seurat: From Moving Points to Depth (CVPR 2025 · similarity 0.86)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 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.
