Research map / Image generation / GANs
Spatial Steerability of GANs via Self-Supervision from Discriminator
Jianyuan Wang, Lalit Bhagat, Ceyuan Yang, Yinghao Xu, Yujun Shen, Hongdong Li, Bolei Zhou
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 SpatialGAN heatmap encoding and alignment | StyleGAN2 · LSUN Bedroom | FID: 4.27 -> 2.72 | Improved |
| add SpatialGAN heatmap encoding and alignment | StyleGAN2 · LSUN Cat | FID: 8.36 -> 6.57 | Improved |
| add SpatialGAN heatmap encoding and alignment | StyleGAN2 · LSUN Church | FID: 3.73 -> 2.86 | Improved |
| add multi-object heatmap sampling (ablation) | SpatialGAN conference version · LSUN Bedroom | Co-move Ratio: 0.32 -> 0.41 | Improved |
| add SELindoor (ablation) | SpatialGAN · LSUN Bedroom | Co-move Ratio: 0.41 -> 0.62 | Improved |
| add indoor self-supervision objective (ablation) | SpatialGAN · LSUN Bedroom | Co-move Ratio: 0.62 -> 0.69 | Improved |
| feed 2D Gaussian noise instead of heatmap (ablation) | SpatialGAN · LSUN Cat | FID: 8.36 -> 8.33 | No clear effect |
| add hierarchical heatmap sampling over non-hierarchical (ablation) | SpatialGAN · LSUN Cat | FID: 7.03 -> 6.57 | Improved |
| add hierarchical heatmap sampling over non-hierarchical (ablation) | SpatialGAN · FFHQ | FID: 3.29 -> 2.91 | Improved |
| flatten heatmap into the latent vector instead of 2D SEL encoding (ablation) | StyleGAN2 · LSUN Cat | FID: 8.36 -> 8.52 | Hurt |
| use SELnorm instead of SELconcat (ablation) | SpatialGAN · LSUN Cat | FID: 6.73 -> 6.57 | Improved |
| use 7x7 kernel instead of 3x3 without coarse heatmap processing (ablation) | SpatialGAN · LSUN Cat | FID: 6.81 -> 6.72 | Improved |
Closest papers by meaning
- On the "steerability" of generative adversarial networks (ICLR 2020 · similarity 0.91)
- Exploiting Spatial Dimensions of Latent in GAN for Real-Time Image Editing (CVPR 2021 · similarity 0.90)
- GAN "Steerability" without optimization (ICLR 2021 · similarity 0.90)
- Low-Rank Subspaces in GANs (NeurIPS 2021 · similarity 0.89)
- PandA: Unsupervised Learning of Parts and Appearances in the Feature Maps of GANs (ICLR 2023 · similarity 0.89)
- WeditGAN: Few-Shot Image Generation via Latent Space Relocation (AAAI 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.
