Research map / Image generation / Image restoration
StyleGAN knows Normal, Depth, Albedo, and More
Anand Bhattad, Daniel McKee, Derek Hoiem, David Forsyth
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 |
|---|---|---|---|
| Use StyleGAN latent offset instead of supervised normal predictor (Omnidata-v1) | StyleGAN · Normals | L1: 0.0501 -> 0.0834 | Hurt |
| Use StyleGAN latent offset instead of Omnidata-v1 on normals | StyleGAN · Normals | Angular error: 0.0750 -> 0.1216 | Hurt |
| Use StyleGAN latent offset instead of Omnidata-v2 for depth | StyleGAN · Depth | L1: 0.1237 -> 0.1019 | Improved |
| Use StyleGAN instead of X-task consistency for depth | StyleGAN · Depth | L1: 0.1390 -> 0.1019 | Improved |
| Use StyleGAN offset instead of DPT for segmentation | StyleGAN · Segmentation mean | mIoU: 87.7 -> 82.0 | Hurt |
| Use StyleGAN offset instead of DPT for segmentation | StyleGAN · Segmentation mean | Accuracy: 98.8 -> 98.1 | Hurt |
| Use StyleGAN offset instead of DPT for lamp segmentation | StyleGAN · lamp | mIoU: 79.3 -> 71.9 | Hurt |
| Use StyleGAN offset instead of DPT for pillow segmentation | StyleGAN · pillow | mIoU: 82.5 -> 76.6 | Hurt |
| Use StyleGAN offset instead of DPT for painting segmentation | StyleGAN · painting | mIoU: 92.9 -> 84.0 | Hurt |
| Use StyleGAN offsets rather than regression networks under relighting | StyleGAN · Normals under 16 lighting conditions | angular change: 8 degrees | Improved |
| Search for an offset that swaps left and right image halves (ablation) | StyleGAN · Control task | qualitative | Hurt |
| Use L1 loss versus task-specific losses (scale-invariant, classification) for offset search (ablation) | StyleGAN · Depth/segmentation | performance | No clear effect |
Closest papers by meaning
- A Survey on Intrinsic Images: Delving Deep into Lambert and Beyond (IJCV 2022 · similarity 0.89)
- Latent Intrinsics Emerge from Training to Relight (NeurIPS 2024 · similarity 0.88)
- StyLitGAN: Image-Based Relighting via Latent Control (CVPR 2024 · similarity 0.86)
- Generative Photography: Scene-Consistent Camera Control for Realistic Text-to-Image Synthesis (CVPR 2025 · similarity 0.85)
- Intrinsic Image Diffusion for Indoor Single-view Material Estimation (CVPR 2024 · similarity 0.84)
- IntrinsiX: High-Quality PBR Generation using Image Priors (NeurIPS 2025 · 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.
