Research map / Image generation / GANs
Aligning and Projecting Images to Class-conditional Generative Networks
Minyoung Huh, Richard Zhang, Jun-Yan Zhu, Sylvain Paris, Aaron Hertzmann
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 BasinCMA instead of ADAM (ablation) | projection optimizer · ImageNet (avg of 18 seeds) | VGG-LPIPS (lower is better): 0.58 -> 0.51 | Improved |
| use CMA instead of L-BFGS (ablation) | projection optimizer · ImageNet (avg of 18 seeds) | VGG-LPIPS (lower is better): 0.61 -> 0.55 | Improved |
| add spatial and color transform and encoder to BasinCMA (ablation) | BasinCMA · ImageNet (avg of 18 seeds) | VGG-LPIPS (lower is better): 0.51 -> 0.47 | Improved |
| add spatial transformation to BasinCMA (ablation) | BasinCMA · ImageNet (avg of 18 seeds) | VGG-LPIPS (lower is better): 0.51 -> 0.48 | Improved |
| add color transformation (brightness) to BasinCMA (ablation) | BasinCMA · ImageNet (avg of 18 seeds) | VGG-LPIPS (lower is better): 0.51 -> 0.50 | No clear effect |
| add transformation to ADAM (ablation) | ADAM · COCO (best of 18 seeds) | VGG-LPIPS (lower is better): 0.56 -> 0.52 | Improved |
| add transformation to BasinCMA (ablation) | BasinCMA · COCO (best of 18 seeds) | VGG-LPIPS (lower is better): 0.53 -> 0.49 | Improved |
| use predicted class instead of random Gaussian class initialization (ablation) | ADAM · ImageNet (best of 18 seeds) | VGG-LPIPS (lower is better): 0.86 -> 0.52 | Improved |
| use ground truth class instead of predicted class (ablation) | ADAM · ImageNet (best of 18 seeds) | VGG-LPIPS (lower is better): 0.52 -> 0.51 | No clear effect |
| use CMA to optimize class vector instead of gradient descent (ablation) | class vector · ImageNet | VGG-LPIPS (lower is better) | No clear effect |
Closest papers by meaning
- Dual Projection Generative Adversarial Networks for Conditional Image Generation (ICCV 2021 · similarity 0.90)
- Conditional Image Synthesis with Auxiliary Classifier GANs (ICML 2017 · similarity 0.90)
- Large Scale GAN Training for High Fidelity Natural Image Synthesis (ICLR 2019 · similarity 0.89)
- Attack Deterministic Conditional Image Generative Models for Diverse and Controllable Generation (AAAI 2024 · similarity 0.87)
- Taming the Tail in Class-Conditional GANs: Knowledge Sharing via Unconditional Training at Lower Resolutions (CVPR 2024 · similarity 0.87)
- SAN: Inducing Metrizability of GAN with Discriminative Normalized Linear Layer (ICLR 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.
