Research map / Image generation / GANs
Learning Layout and Style Reconfigurable GANs for Controllable Image Synthesis
Wei Sun, Tianfu Wu
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 |
|---|---|---|---|
| add object head classifier in discriminator (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | FID: 57.06 -> 24.76 | Improved |
| add layout-to-mask component in generator (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | FID: 28.51 -> 24.76 | Improved |
| add mask refinement in generator (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | FID: 26.11 -> 24.76 | Improved |
| add image head classifier in discriminator (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | FID: 25.96 -> 24.76 | Improved |
| add all mask refinement stages m1..m5 (ablation) | LostGAN-V2 · COCO-Stuff 256x256 | FID: 48.54 -> 42.55 | Improved |
| add only first refinement stage m1 (coarse resolution) (ablation) | LostGAN-V2 · COCO-Stuff 256x256 | FID: 48.54 -> 63.96 | Hurt |
| train with pseudo-box labels on second half of data (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | FID: 32.77 -> 25.87 | Improved |
| use full box annotations instead of half (ablation) | LostGAN-V2 · COCO-Stuff 128x128 | DS: 0.46 -> 0.45 | No clear effect |
| LostGAN-V1 vs Layout2Im at 64x64 | LostGAN-V1 · COCO-Stuff 64x64 | DS: 0.15 -> 0.35 | Improved |
| LostGAN-V2 vs LostGAN-V1 at 128x128 | LostGAN-V2 · Visual Genome 128x128 | IS: 11.10 -> 10.71 | Hurt |
| LostGAN-V2 vs Grid2Im at 256x256 | LostGAN-V2 · COCO-Stuff 256x256 | DS: 0.34 -> 0.55 | Improved |
| raise resolution from 256 to 512 | LostGAN-V2 · COCO-Stuff | IS, FID, DS: FID 42.55 -> 51.99; DS 0.55 -> 0.65 | Mixed |
Closest papers by meaning
- Image Synthesis From Reconfigurable Layout and Style (ICCV 2019 · similarity 0.95)
- Interactive Image Synthesis With Panoptic Layout Generation (CVPR 2022 · similarity 0.90)
- Context-Aware Layout to Image Generation With Enhanced Object Appearance (CVPR 2021 · similarity 0.90)
- Image Generation From Layout (CVPR 2019 · similarity 0.89)
- Cluster-Guided Image Synthesis With Unconditional Models (CVPR 2022 · similarity 0.89)
- Learning to Predict Layout-to-image Conditional Convolutions for Semantic Image Synthesis (NeurIPS 2019 · similarity 0.89)
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.
