Research map / 3D vision / Depth and stereo
Single Image Depth Prediction Made Better: A Multivariate Gaussian Take
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| model per-pixel depth with a multivariate Gaussian and learned low-rank covariance | Swin-L SIDP model · NYU Depth V2 test | SILog: 9.102 -> 8.323 | Helped |
| model per-pixel depth with a multivariate Gaussian and learned low-rank covariance | Swin-L SIDP model · KITTI official test | SILog: 10.39 -> 9.93 | Helped |
| model per-pixel depth with a multivariate Gaussian and learned low-rank covariance | Swin-L SIDP model · KITTI Eigen test | SILog: 6.986 -> 6.757 | Helped |
| model per-pixel depth with a multivariate Gaussian and learned low-rank covariance | Swin-L SIDP model · SUN RGB-D test, trained on NYU without fine-tuning | SILog: 13.652 -> 11.985 | Helped |
| replace NLL with L2 loss (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | SILog: 8.323 -> 8.912 | Hurt |
| replace NLL with SI loss (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | SILog: 8.323 -> 8.762 | Hurt |
| replace NLL with gradient loss (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | SILog: 8.323 -> 8.886 | Hurt |
| replace NLL with VNL loss (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | SILog: 8.323 -> 8.543 | Hurt |
| use our NLL loss instead of the original loss (ablation) | NeWCRFs · NYU Depth V2 test | SILog: 9.102 -> 8.619 | Helped |
| increase the covariance factor rank (ablation) | proposed Swin-L SIDP model · NYU Depth V2 | depth prediction accuracy | Helped |
| omit K-Decoder at test time (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | SILog: 8.323 -> 8.323 | No clear effect |
| omit K-Decoder at test time (ablation) | proposed Swin-L SIDP model · NYU Depth V2 test | FPS: 8.445 -> 9.909 | Helped |
Closest papers by meaning
- VA-DepthNet: A Variational Approach to Single Image Depth Prediction (ICLR 2023 · similarity 0.90)
- Dense Depth Posterior (DDP) From Single Image and Sparse Range (CVPR 2019 · similarity 0.86)
- Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025 · similarity 0.86)
- P3Depth: Monocular Depth Estimation With a Piecewise Planarity Prior (CVPR 2022 · similarity 0.85)
- Zero-shot Depth Completion via Test-time Alignment with Affine-invariant Depth Prior (AAAI 2025 · similarity 0.85)
- Depth Prompting for Sensor-Agnostic Depth Estimation (CVPR 2024 · similarity 0.84)
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.