Research map / Image generation / GANs
A Multi-Resolution Framework for U-Nets with Applications to Hierarchical VAEs
Fabian Falck, Christopher Williams, Dominic Danks, George Deligiannidis, Christopher Yau, Christopher C. Holmes, Arnaud Doucet, Matthew Willetts
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 |
|---|---|---|---|
| add weight sharing across time to ResNet blocks | VDVAE* · MNIST | NLL (nats): 80.14 -> 79.98 | No clear effect |
| add weight sharing across time to ResNet blocks | VDVAE* · CIFAR10 | NLL (bits/dim): 2.87 -> 2.88 | No clear effect |
| add weight sharing with 39m parameters | VDVAE* · CIFAR10 | NLL (bits/dim): 2.87 -> 2.83 | Improved |
| add weight sharing across time to ResNet blocks | VDVAE* · ImageNet32 | NLL (bits/dim): 3.67 -> 3.68 | No clear effect |
| add weight sharing with 85m parameters | VDVAE* · ImageNet32 | NLL (bits/dim): 3.67 -> 3.65 | Improved |
| add weight sharing across time to ResNet blocks | VDVAE* · CelebA 64x64 | NLL (bits/dim): 2.02 -> 2.02 | No clear effect |
| repeat a single parameterised block more times (r2 to r20) with weight sharing (ablation) | WS-VDVAE · MNIST and CIFAR10 validation | NLL | Improved |
| normalise residual state by its norm at every block (ablation) | HVAE · MNIST | NLL (nats): 81.69 -> 464.68 | Hurt |
| normalise residual state by its norm at every block (ablation) | HVAE · CIFAR10 | NLL (bits/dim): 2.93 -> 6.80 | Hurt |
| normalise residual state by its norm at every block (ablation) | HVAE · ImageNet | NLL (bits/dim): 3.68 -> 6.76 | Hurt |
| increase number of latent variables / stochastic depth (ablation) | HVAE · ablation, multiple datasets | NLL | No clear effect |
| add Fourier features (ablation) | HVAE · ablation, multiple datasets | NLL | No clear effect |
Closest papers by meaning
- U-Nets as Belief Propagation: Efficient Classification, Denoising, and Diffusion in Generative Hierarchical Models (ICLR 2025 · similarity 0.86)
- UDT: Reconciling U-Nets and Diffusion Transformers with Data-Adaptive Token Reduction (NeurIPS 2026 · similarity 0.84)
- U-DiTs: Downsample Tokens in U-Shaped Diffusion Transformers (NeurIPS 2024 · similarity 0.84)
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 2024 · similarity 0.83)
- FreeU: Free Lunch in Diffusion U-Net (CVPR 2024 · similarity 0.83)
- U-REPA: Aligning Diffusion U-Nets to ViTs (NeurIPS 2025 · similarity 0.83)
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.
