Research map / 3D vision / 3D generation
BlobGAN: Spatially Disentangled Scene Representations
Dave Epstein, Taesung Park, Richard Zhang, Eli Shechtman, Alexei A. Efros
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 BlobGAN instead of StyleGAN2 | image generator · LSUN bedrooms | FID: 3.85 -> 3.43 | Improved |
| use BlobGAN instead of StyleGAN2 | image generator · LSUN Kitchen+Living+Dining | FID: 4.63 -> 4.41 | Improved |
| use BlobGAN instead of StyleGAN2 | image generator · LSUN Conference | FID: 6.21 -> 6.94 | Hurt |
| condition target scene on layout via scene auto-complete instead of random target | BlobGAN edits · LSUN bedrooms Layout->Styles | FID: 8.10 -> 4.59 | Improved |
| condition target scene on window instead of random target | BlobGAN edits · LSUN bedrooms Window->Room | FID: 6.41 -> 4.75 | Improved |
| condition target scene on bed instead of random target | BlobGAN edits · LSUN bedrooms Bed->Room | FID: 10.88 -> 7.12 | Improved |
| condition target scene on painting instead of random target | BlobGAN edits · LSUN bedrooms Painting->Room | FID: 6.31 -> 4.58 | Improved |
| swap styles in BlobGAN instead of StyleGAN layer 4 | style swapping · LSUN bedrooms | layout consistency: 62.9 -> 63.6 | Improved |
| swap styles in BlobGAN instead of StyleGAN layer 3 | style swapping · LSUN bedrooms | FID: 5.04 -> 5.06 | No clear effect |
| use k=20 blobs instead of 10 (ablation) | BlobGAN · LSUN bedrooms | FID: 3.73 | Hurt |
| remove blob jitter (ablation) | BlobGAN · LSUN bedrooms | FID | No clear effect |
| add style mixing during training (ablation) | BlobGAN · LSUN | training quality | Hurt |
Closest papers by meaning
- Factoring Shape, Pose, and Layout From the 2D Image of a 3D Scene (CVPR 2018 · similarity 0.87)
- Fast and Flexible Indoor Scene Synthesis via Deep Convolutional Generative Models (CVPR 2019 · similarity 0.87)
- BlockGAN: Learning 3D Object-aware Scene Representations from Unlabelled Images (NeurIPS 2020 · similarity 0.87)
- Learning Object Context for Novel-View Scene Layout Generation (CVPR 2022 · similarity 0.87)
- RELATE: Physically Plausible Multi-Object Scene Synthesis Using Structured Latent Spaces (NeurIPS 2020 · similarity 0.87)
- Disentangled 3D Scene Generation with Layout Learning (ICML 2024 · similarity 0.86)
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.
