Research map / Image generation / GANs
High-Fidelity Synthesis with Disentangled Representation
Wonkwang Lee, Donggyun Kim, Seunghoon Hong, Honglak Lee
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 |
|---|---|---|---|
| train encoder and generator in two successive steps instead of end-to-end (ablation) | ID-GAN · dSprites | FactorVAE metric: 0.50 -> 0.65 | Improved |
| train encoder and generator in two successive steps instead of end-to-end (ablation) | ID-GAN · dSprites | MIG: 0.13 -> 0.28 | Improved |
| train encoder and generator in two successive steps instead of end-to-end (ablation) | ID-GAN · dSprites | FID: 3.18 -> 2.00 | Improved |
| use two-step ID-GAN instead of joint VAE-GAN (ablation) | beta-VAE · dSprites | MIG: 0.13 -> 0.28 | Improved |
| add adversarial loss to the encoder update (VAE-GAN vs beta-VAE) (ablation) | beta-VAE · dSprites | FactorVAE metric: 0.65 -> 0.46 | Hurt |
| add adversarial loss to the encoder update (VAE-GAN vs beta-VAE) (ablation) | beta-VAE · dSprites | FID: 37.75 -> 33.54 | Improved |
| add distillation regularizer R_ours to generator training (ablation) | ID-GAN · CelebA 128x128 | R_ours (alignment): -65.84 -> -10.25 | Improved |
| add distillation regularizer R_ours to generator training (ablation) | ID-GAN · CelebA 128x128 | GILBO: -20.40 -> -0.19 | Improved |
| add distillation regularizer R_ours to generator training (ablation) | ID-GAN · CelebA 128x128 | FID: 5.75 -> 6.61 | No clear effect |
| replace implicit cGAN alignment with explicit distillation | two-step conditional generation · CelebA 128x128 | GILBO: -7.57 -> -0.19 | Improved |
| replace beta-VAE with ID-GAN on beta-VAE code | beta-VAE · CelebA (64x64) | FID: 166.01 -> 4.08 | Improved |
| replace InfoGAN information maximization with distilled VAE encoder | InfoGAN · 3D Chairs (64x64) | FID: 60.45 -> 25.44 | Improved |
Closest papers by meaning
- IB-GAN: Disentangled Representation Learning with Information Bottleneck Generative Adversarial Networks (AAAI 2021 · similarity 0.92)
- Learning Disentangled Representation by Exploiting Pretrained Generative Models: A Contrastive Learning View (ICLR 2022 · similarity 0.90)
- InfoGAN-CR and ModelCentrality: Self-supervised Model Training and Selection for Disentangling GANs (ICML 2020 · similarity 0.89)
- Disentangled Image Generation Through Structured Noise Injection (CVPR 2020 · similarity 0.89)
- Distilling Representations from GAN Generator via Squeeze and Span (NeurIPS 2022 · similarity 0.88)
- Transferring disentangled representations: bridging the gap between synthetic and real images (NeurIPS 2024 · similarity 0.88)
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.
