Research map / Image generation / Autoregressive and latent generation
Catch Missing Details: Image Reconstruction With Frequency Augmented Variational Autoencoder
Read the paperCheck 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 |
|---|---|---|---|
| Add FCM and DSL | VQ-GAN-style VQ-VAE · ImageNet | rFID: 1.06 -> 0.40 | Helped |
| Add FCM and DSL | VQ-GAN-style VQ-VAE · FFHQ | rFID: 5.33 -> 4.98 | Helped |
| Use FA-VAE instead of VQ-GAN at f=16 | ImageNet-pretrained reconstruction model · CelebA-HQ zero-shot | rFID: 8.62 -> 6.52 | Helped |
| Use FA-VAE instead of VQ-GAN at f=16 | ImageNet-pretrained reconstruction model · FFHQ zero-shot | PSNR: 22.68 -> 21.95 | Hurt |
| Use FA-VAE instead of VQ-GAN at f=4 | ImageNet-pretrained reconstruction model · CelebA-HQ zero-shot | PSNR: 31.20 -> 31.39 | Helped |
| Use FA-VAE instead of VQ-GAN at f=4 | ImageNet-pretrained reconstruction model · FFHQ zero-shot | rFID: 3.06 -> 1.46 | Helped |
| Use CAT with all text-token embeddings and cross-attention at each generation step | GPT2 autoregressive image-token generator · CelebA-HQ-MM | FID: 12.54 -> 10.23 | Helped |
| Add image-level FFL without FCM (ablation) | VQ-GAN · CelebA-HQ validation | rFID: 10.12 -> 30.65 | Hurt |
| Add FCM convolution modules without kernel-based spectrum loss (ablation) | VQ-GAN · CelebA-HQ validation | rFID: 10.12 -> 7.04 | Helped |
| Add image-level SL to convolutional FCMs (ablation) | FA-VAE · CelebA-HQ validation | rFID: 7.04 -> 7.02 | Helped |
| Use dynamic rather than fixed spectrum weights across decoder blocks (ablation) | FA-VAE with FCM · CelebA-HQ validation | rFID: 7.39 -> 5.90 | Helped |
| Use residual FCM rather than convolutional FCM (ablation) | FA-VAE with DSL · CelebA-HQ validation | rFID: 5.90 -> 6.44 | Hurt |
Closest papers by meaning
- DreamUHD: Frequency Enhanced Variational Autoencoder for Ultra-High-Definition Image Restoration (AAAI 2025 · similarity 0.88)
- SSC-VAE: Structured Sparse Coding Based Variational Autoencoder for Detail Preserved Image Reconstruction (AAAI 2025 · similarity 0.86)
- SeNM-VAE: Semi-Supervised Noise Modeling with Hierarchical Variational Autoencoder (CVPR 2024 · similarity 0.85)
- Boosting ViT-based MRI Reconstruction from the Perspectives of Frequency Modulation, Spatial Purification, and Scale Diversification (AAAI 2025 · similarity 0.84)
- Latent Harmony: Synergistic Unified UHD Image Restoration via Latent Space Regularization and Controllable Refinement (NeurIPS 2025 · similarity 0.84)
- Epsilon-VAE: Denoising as Visual Decoding (ICML 2025 · similarity 0.84)
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.