Research map / Image generation / GANs
Inducing Optimal Attribute Representations for Conditional GANs
Binod Bhattarai, Tae-Kyun Kim
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 |
|---|---|---|---|
| switch conditioning from target (Std) to target-minus-source (Diff) one-hot (ablation) | StarGAN · CelebA TARR avg | TARR (%): 67.2% -> 74.5% | Improved |
| replace one-hot with GCN-induced attribute representations (ablation) | StarGAN (Diff) · CelebA TARR avg | TARR (%): 74.5 -> 79.4 | Improved |
| add MTL at discriminator to gcn-reprs (ablation) | StarGAN · CelebA TARR avg | TARR (%): 79.4 -> 81.0 | Improved |
| train GCN and GAN end-to-end instead of multi-stage (ablation) | StarGAN + gcn-reprs + MTL · CelebA TARR avg | TARR (%): 81.0 -> 83.4 | Improved |
| add MTL at discriminator (13 attributes) (ablation) | StarGAN one-hot Std · CelebA TARR avg | TARR (%): 67.2 -> 68.3 | Improved |
| replace one-hot with latent-reprs (ablation) | StarGAN (Diff) · CelebA TARR avg | TARR (%): 74.5 -> 71.4 | Hurt |
| replace one-hot with attrbs-weights (ablation) | StarGAN (Diff) · CelebA TARR avg | TARR (%): 74.5 -> 76.0 | Improved |
| replace one-hot with co-occurrence vectors (ablation) | StarGAN (Diff, 5 attributes) · CelebA TARR avg (5 attributes) | TARR (%): 80.2 -> 78.6 | Hurt |
| replace one-hot with word2vec (ablation) | StarGAN (Diff, 5 attributes) · CelebA TARR avg (5 attributes) | TARR (%): 80.2 -> 81.3 | Improved |
| replace one-hot with gcn-reprs (ablation) | StarGAN (Diff, 5 attributes) · CelebA TARR avg (5 attributes) | TARR (%): 80.2 -> 84.0 | Improved |
| apply gcn-reprs + MTL + End2End | STGAN · CelebA TARR avg | TARR (%): 86.1 -> 89.1 | Improved |
| apply gcn-reprs + MTL + End2End | AttGAN · CelebA TARR avg | TARR (%): 83 -> 90.3 | Improved |
Closest papers by meaning
- RoCGAN: Robust Conditional GAN (IJCV 2020 · similarity 0.91)
- Dual Projection Generative Adversarial Networks for Conditional Image Generation (ICCV 2021 · similarity 0.90)
- Continuous Conditional Generative Adversarial Networks: Novel Empirical Losses and Label Input Mechanisms (TPAMI 2022 · similarity 0.90)
- cGANs with Projection Discriminator (ICLR 2018 · similarity 0.90)
- A Unified View of cGANs with and without Classifiers (NeurIPS 2021 · similarity 0.90)
- SDGAN: Disentangling Semantic Manipulation for Facial Attribute Editing (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.
