Research map / 3D vision / Depth and stereo
Omnidata: A Scalable Pipeline for Making Multi-Task Mid-Level Vision Datasets From 3D Scans
Ainaz Eftekhar, Alexander Sax, Jitendra Malik, Amir Zamir
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 |
|---|---|---|---|
| train depth model on Omnidata starter data instead of MiDaS data | DPT-Hybrid · NYU zero-shot depth | L1 error: 0.3838 -> 0.2878 | Improved |
| train depth model on Omnidata starter data instead of MiDaS data | DPT-Hybrid · OASIS zero-shot depth | L1 error: 0.8057 -> 0.7901 | Improved |
| train normals UNet on Omnidata instead of OASIS-trained Hourglass | UNet · OASIS surface normals | AUC_p: 0.5786 -> 0.6758 | Improved |
| train normals UNet on Omnidata instead of OASIS-trained Hourglass | UNet · OASIS surface normals | mean angular error: 23.91 -> 24.87 | Hurt |
| train UNet on Omnidata instead of SunCG | UNet · OASIS surface normals | mean angular error: 35.42 -> 24.87 | Improved |
| add normals as extra input cue (ablation) | HRNet-18 · Semantic segmentation, Replica to Hypersim (GT cues) | cross-entropy: 5.87 -> 4.47 | Improved |
| add 2D edges, z-depth and 3D keypoints cues to the input (ablation) | HRNet-18 · Semantic segmentation, Hypersim (GT cues) | cross-entropy: 4.47 -> 3.86 | Improved |
| add 3D edges cue to the input (ablation) | HRNet-18 · Semantic segmentation, Taskonomy (GT cues) | cross-entropy: 6.12 -> 6.75 | Hurt |
| train on all starter data instead of a single component dataset | Surface normal UNet · Inter-dataset transfer | harmonic mean L1: 8.24 -> 6.11 | Improved |
| use multi-task learning methods (MTAN, cross-stitch) over single task | MTL methods · Omnidata semantic segmentation | IoU: 85.12 -> 83.00 | Hurt |
| apply MTAN to MTL baseline | MTL baseline · Omnidata semantic segmentation | IoU: 81.82 -> 83.00 | Improved |
Later papers that compared with this method
3 later accepted papers report Omnidata v2 as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Rethinking Inductive Biases for Surface Normal Estimation (CVPR 2024 · better on 2 of 3 shared results)
- GeoWizard: Unleashing the Diffusion Priors for 3D Geometry Estimation from a Single Image (ECCV 2024 · better on 1 of 1 shared result)
- Unsupervised Volumetric Animation (CVPR 2023 · better on 2 of 2 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
- OmniDepth: Dense Depth Estimation for Indoors Spherical Panoramas (ECCV 2018 · similarity 0.89)
- HELVIPAD: A Real-World Dataset for Omnidirectional Stereo Depth Estimation (CVPR 2025 · similarity 0.87)
- Depth Anything: Unleashing the Power of Large-Scale Unlabeled Data (CVPR 2024 · similarity 0.86)
- A Simple yet Universal Framework for Depth Completion (NeurIPS 2024 · similarity 0.86)
- UniK3D: Universal Camera Monocular 3D Estimation (CVPR 2025 · similarity 0.86)
- Depth2Pose: A Pose-Based Benchmark for Monocular Depth Estimation without Ground-Truth Depth (NeurIPS 2026 · similarity 0.85)
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.
