Research map / Image generation / GANs
Hierarchical Semantic Regularization of Latent Spaces in StyleGANs
Tejan Karmali, Rishubh Parihar, Susmit Agrawal, Harsh Rangwani, Varun Jampani, Maneesh Singh, R. Venkatesh Babu
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 HSR | StyleGAN2 · FFHQ-140k | PPL: 175.09 -> 144.59 | Improved |
| add HSR | StyleGAN2-ADA · FFHQ-140k | FID: 4.30 -> 5.26 | Hurt |
| add HSR | StyleGAN2-ADA · FFHQ-10k | PPL: 164.21 -> 126.35 | Improved |
| add HSR | StyleGAN2-ADA · FFHQ-10k | FID: 7.16 -> 8.08 | Hurt |
| add HSR | StyleGAN2-ADA · FFHQ-1k | PPL: 98.79 -> 90.39 | Improved |
| use pretrained ViT DINO features instead of randomly initialised ViT (ablation) | SG2-ADA+HSR · AnimalFace-Dog | PPL: 56.97 -> 48.02 | Improved |
| use ResNet50 DINO as extractor (ablation) | SG2-ADA+HSR · AnimalFace-Dog | FID: 53.28 -> 54.33 | Hurt |
| supervise only low-level blocks (ablation) | SG2-ADA+HSR · AnimalFace-Dog | FID/PPL: FID 53.28 -> 53.66, PPL 59.27 -> 51.77 | Mixed |
| supervise all levels instead of only low-level (ablation) | SG2-ADA+HSR · AnimalFace-Dog | PPL: 51.77 -> 48.02 | Improved |
| add PLR to baseline (ablation) | StyleGAN2-ADA · AnimalFace-Dog | PPL: 75.63 -> 59.27 | Improved |
| add HSR on top of PLR (ablation) | StyleGAN2-ADA · AnimalFace-Dog | PPL: 59.27 -> 48.02 | Improved |
| add HSR under truncation 0.7 | StyleGAN2 · FFHQ-140k (truncation) | FID: 21.46 -> 16.94 | Improved |
Closest papers by meaning
- Analyzing and Improving the Image Quality of StyleGAN (CVPR 2020 · similarity 0.91)
- Editing in Style: Uncovering the Local Semantics of GANs (CVPR 2020 · similarity 0.91)
- Semantic Hierarchy Emerges in Deep Generative Representations for Scene Synthesis (IJCV 2021 · similarity 0.90)
- Style Quantization for Data-Efficient GAN Training (CVPR 2025 · similarity 0.90)
- Navigating the GAN Parameter Space for Semantic Image Editing (CVPR 2021 · similarity 0.90)
- Latent Transformations via NeuralODEs for GAN-Based Image Editing (ICCV 2021 · 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.
