Research map / Image generation / Autoregressive and latent generation
Dual Contradistinctive Generative Autoencoder
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 |
|---|---|---|---|
| combine instance-level contrastive loss with set-level GAN loss (ablation) | VAE on CIFAR-10 · CIFAR-10 sampling | FID: 240.4 -> 17.9 | Helped |
| add instance-level contrastive loss (ablation) | VAE/GAN on CIFAR-10 · CIFAR-10 sampling | FID: 39.8 -> 17.9 | Helped |
| add set-level GAN loss (ablation) | VAE-Contrastive on CIFAR-10 · CIFAR-10 sampling | FID: 240.4 -> 17.9 | Helped |
| add instance-level contrastive loss and set-level GAN loss (ablation) | VAE on CIFAR-10 · CIFAR-10 | Perceptual Distance: 65.8 -> 52.9 | Helped |
| replace direct pixel reconstruction objective with instance-level contrastive loss (ablation) | VAE on CIFAR-10 · CIFAR-10 | Pixel Distance: 21.8 -> 45.9 | Hurt |
| remove set-level GAN loss (ablation) | DC-VAE on CIFAR-10 · CIFAR-10 sampling | FID: 17.9 -> 240.4 | Hurt |
| remove instance-level contrastive loss (ablation) | DC-VAE on CIFAR-10 · CIFAR-10 sampling | FID: 17.9 -> 39.8 | Hurt |
| increase the number of contrastive negative samples (ablation) | DC-VAE · CIFAR-10 test set | reconstruction error | Helped |
| use both contradistinctive losses rather than either loss alone (ablation) | DC-VAE on CIFAR-10 · CIFAR-10 sampling | IS: 1.8 -> 8.2 | Helped |
| use instance-level contrastive loss instead of feature reconstruction loss (ablation) | VAE/GAN on CIFAR-10 · CIFAR-10 | Perceptual Distance: 57.2 -> 52.9 | Helped |
| use DC-VAE rather than Progressive GAN | image synthesis on LSUN Bedrooms · LSUN Bedrooms sampling | FID: 8.3 -> 14.3 | Hurt |
| use DC-VAE rather than StyleALAE | CelebA-HQ generation · CelebA-HQ 256×256 | FID: 19.21 -> 15.81 | Helped |
Closest papers by meaning
- Generating Diverse High-Fidelity Images with VQ-VAE-2 (NeurIPS 2019 · similarity 0.88)
- Explicitly Minimizing the Blur Error of Variational Autoencoders (ICLR 2023 · similarity 0.87)
- From Variational to Deterministic Autoencoders (ICLR 2020 · similarity 0.87)
- DC-SAE: Deep Compression Semantic Autoencoder for Faster Diffusion Convergence (NeurIPS 2026 · similarity 0.86)
- Epsilon-VAE: Denoising as Visual Decoding (ICML 2025 · similarity 0.86)
- EQ-VAE: Equivariance Regularized Latent Space for Improved Generative Image Modeling (ICML 2025 · similarity 0.85)
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.