Research map / Image generation / GANs
Generative Hierarchical Features From Synthesizing Images
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Y-space layer-wise style codes instead of W-space codes (ablation) | GH-Feat encoder · FF-HQ reconstruction | MSE: 0.0601 -> 0.0464 | Helped |
| Use Y-space layer-wise style codes instead of W-space codes (ablation) | GH-Feat encoder · FF-HQ reconstruction | SSIM: 0.540 -> 0.558 | Helped |
| Use Y-space layer-wise style codes instead of W-space codes (ablation) | GH-Feat encoder · FF-HQ reconstruction | FID: 22.24 -> 18.48 | Helped |
| Add the Spatial Alignment Module (ablation) | GH-Feat encoder using Y-space codes · FF-HQ reconstruction | MSE: 0.0502 -> 0.0464 | Helped |
| Add the Spatial Alignment Module (ablation) | GH-Feat encoder using Y-space codes · FF-HQ reconstruction | SSIM: 0.550 -> 0.558 | Helped |
| Add the Spatial Alignment Module (ablation) | GH-Feat encoder using Y-space codes · FF-HQ reconstruction | FID: 19.06 -> 18.48 | Helped |
| Reuse a well-trained StyleGAN generator instead of training the generator jointly from scratch (ablation) | GH-Feat encoder · FF-HQ reconstruction | MSE: 0.429 -> 0.0464 | Helped |
| Reuse a well-trained StyleGAN generator instead of training the generator jointly from scratch (ablation) | GH-Feat encoder · FF-HQ reconstruction | SSIM: 0.301 -> 0.558 | Helped |
| Reuse a well-trained StyleGAN generator instead of training the generator jointly from scratch (ablation) | GH-Feat encoder · FF-HQ reconstruction | FID: 46.20 -> 18.48 | Helped |
| Use voting across levels rather than grouping multi-level features | GH-Feat face verification · LFW | accuracy: 60.1 -> 69.7 | Helped |
| Use GH-Feat rather than ALAE features | Unsupervised MNIST digit recognition · MNIST test | Top-1 accuracy: 97.61 -> 99.06 | Helped |
| Use GH-Feat rather than MoCo features | Linear image classifier · ImageNet validation | Top-1 accuracy: 60.6 -> 51.1 | Hurt |
Closest papers by meaning
- Ensembling With Deep Generative Views (CVPR 2021 · similarity 0.88)
- Revisiting the Evaluation of Image Synthesis with GANs (NeurIPS 2023 · similarity 0.88)
- Transformation GAN for Unsupervised Image Synthesis and Representation Learning (CVPR 2020 · similarity 0.87)
- Large Scale Adversarial Representation Learning (NeurIPS 2019 · similarity 0.87)
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.87)
- MSG-GAN: Multi-Scale Gradients for Generative Adversarial Networks (CVPR 2020 · 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.