Research map / 3D vision / Depth and stereo
Unlocking the Power of Critical Factors for 3D Visual Geometry Estimation
Guangkai Xu, Hua Geng, Huanyi Zheng, Songyi Yin, Yanlong Sun, Hao Chen, Chunhua Shen
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 |
|---|---|---|---|
| scale training data from Data1 to Data3 (ablation) | VGGT finetuning · KITTI camera pose | ATE: 1.411 -> 0.937 | Improved |
| add noisy datasets (Data3 over Data2) (ablation) | VGGT finetuning · KITTI camera pose | ATE: 1.267 -> 0.937 | Improved |
| add spatial gradient loss (ablation) | L_reg(W_inv) · KITTI point cloud | Chamfer L1: 0.254 -> 0.270 | Hurt |
| replace learnable confidence weighting with fixed inverse-depth weight (ablation) | L_reg + L_conf · KITTI point cloud | Chamfer L1: 0.270 -> 0.254 | Improved |
| add confidence loss L_conf with learnable weights (ablation) | VGGT loss · KITTI camera pose | ATE: 0.937 -> 1.059 (with L_conf, without L_sg: 1.059) | Hurt |
| add temporal gradient loss (ablation) | L_reg(W_inv) · KITTI camera pose | ATE: 0.866 -> 1.270 | Hurt |
| add per-frame scale-shift alignment loss L_F (ablation) | L_reg(W_inv) · KITTI point cloud | Chamfer L1: 0.254 -> 0.245 | Improved |
| add local region alignment loss L_S (ablation) | L_reg(W_inv) + L_F · KITTI point cloud | Chamfer L1: 0.245 -> 0.255 | Hurt |
| add geometric consistency loss (ablation) | L_reg(W_inv) + L_F · KITTI camera pose | ATE: 1.042 -> 0.919 | Improved |
| add efficient cross-attention high-resolution fusion (ablation) | CARVE · KITTI point cloud | Chamfer L1: 0.249 -> 0.238 | Improved |
| add efficient high-resolution fusion instead of direct high-resolution VGGT input (max 100 frames) (ablation) | CARVE · KITTI camera pose | ATE: 0.355 -> 0.289 | Improved |
| use efficient high-resolution fusion instead of direct 1036x1036 input | VGGT · Inference speed, 32 frames | FPS: 2.54 -> 15.26 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- VGGT — VGGT: Visual Geometry Grounded Transformer (CVPR 2025) · better on 9 of 10 shared results
- MoGe — MoGe: Unlocking Accurate Monocular Geometry Estimation for Open-Domain Images with Optimal Training Supervision (CVPR 2025) · better on 0 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
- Robust Geometry-Preserving Depth Estimation Using Differentiable Rendering (ICCV 2023 · similarity 0.88)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.87)
- PatchFusion: An End-to-End Tile-Based Framework for High-Resolution Monocular Metric Depth Estimation (CVPR 2024 · similarity 0.87)
- SurGe: Improved Surface Geometry in Point Maps (NeurIPS 2026 · similarity 0.87)
- Not All Tasks Quantize Equally: Fisher-Guided Quantization for Visual Geometry Transformer (NeurIPS 2026 · similarity 0.87)
- Seurat: From Moving Points to Depth (CVPR 2025 · similarity 0.87)
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.
