Research map / 3D vision / Depth and stereo
DAViD: Data-efficient and Accurate Vision Models from Synthetic Data
Fatemeh Saleh, Sadegh Aliakbarian, Charlie Hewitt, Lohit Petikam, Xian Xiao, Antonio Criminisi, Thomas J. Cashman, Tadas Baltrusaitis
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 on SynthHuman instead of THuman2.0 renders (ablation) | Large depth model · Goliath depth | RMSE: 0.495 -> 0.253 | Improved |
| train on SynthHuman instead of RenderPeople renders (ablation) | Large depth model · Goliath depth | RMSE: 0.278 -> 0.253 | Improved |
| increase training set from 60K to 300K images (ablation) | ViT-Base depth · Goliath depth | RMSE: 0.324 -> 0.278 | Improved |
| increase training set from 60K to 300K images (ablation) | ViT-Base depth · Hi4D depth | RMSE: 0.101 -> 0.085 | Improved |
| scale backbone from ViT-Small to ViT-Large (ablation) | depth model · Goliath depth | RMSE: 0.310 -> 0.253 | Improved |
| scale backbone from ViT-Base to ViT-Large (ablation) | depth model · Hi4D depth | RMSE: 0.085 -> 0.072 | Improved |
| single multi-task Large model instead of three separate Large models (ablation) | Large model · Goliath depth | RMSE: 0.253 -> 0.270 | Hurt |
| single multi-task Large model instead of three separate Large models (ablation) | Large model · PPM-100 matting | SAD: 78.17 -> 66.08 | Improved |
| multi-task Large (0.35B) instead of three Base models (3x0.12B) (ablation) | Large model · Goliath depth | RMSE: 0.278 -> 0.270 | Improved |
| blur the normal prediction output (ablation) | Ours-Large normals · Goliath-FullBody | mean angular error: 14.60 -> 14.52 | Improved |
| blur the normal prediction output (ablation) | Sapiens-0.3B normals · Goliath-FullBody | mean angular error: 15.72 -> 15.69 | No clear effect |
| fine-tune SynthHuman model on P3M-train | Ours matting · P3M-500-NP | SAD: 14.83 -> 9.12 | 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.
- Sapiens-2B — Sapiens: Foundation for Human Vision Models (ECCV 2024) · better on 5 of 8 shared results
- Depth-Pro — Depth Pro: Sharp Monocular Metric Depth in Less Than a Second (ICLR 2025) · 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
- When Pretty Isn't Useful: Investigating Why Modern Text-to-Image Models Fail as Reliable Training Data Generators (CVPR 2026 · similarity 0.89)
- Real-Fake: Effective Training Data Synthesis Through Distribution Matching (ICLR 2024 · similarity 0.89)
- The Unmet Promise of Synthetic Training Images: Using Retrieved Real Images Performs Better (NeurIPS 2024 · similarity 0.88)
- Scaling Laws of Synthetic Images for Model Training ... for Now (CVPR 2024 · similarity 0.87)
- Improving the Scaling Laws of Synthetic Data with Deliberate Practice (ICML 2025 · similarity 0.87)
- BLURD: Benchmarking and Learning using a Unified Rendering and Diffusion Model (NeurIPS 2024 · 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.
