Research map / Architectures and efficiency / CNN design
Scale-Space Hypernetworks for Efficient Biomedical Image Analysis
Jose Javier Gonzalez Ortiz, John Guttag, Adrian V Dalca
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 |
|---|---|---|---|
| replace Fixed U-Net with SSHN hypernetwork | U-Net · CAMUS test | Dice (%): 83.5 -> 86.2 | Improved |
| replace Fixed U-Net with SSHN hypernetwork | U-Net · OASIS test | Dice (%): 89.1 -> 90.7 | Improved |
| replace Fixed U-Net with SSHN hypernetwork | U-Net · PanDental test | Dice (%): 88.9 -> 89.9 | Improved |
| replace Fixed U-Net with SSHN hypernetwork | U-Net · WBC test | Dice (%): 90.9 -> 92.7 | Improved |
| use stochastic rescaling factor during training of a plain U-Net instead of fixed | U-Net · OASIS test | Dice (%): 89.1 -> 84.8 | Hurt |
| add FiLM conditioning on stochastically sampled rescaling factor instead of fixed | U-Net · OASIS test | Dice (%): 89.1 -> 87.2 | Hurt |
| stochastic rescaling vs fixed on CAMUS | U-Net · CAMUS test | Dice (%): 83.5 -> 83.3 | No clear effect |
| widen the training prior of phi from r=0 (constant 0.5) to wider uniform range (ablation) | SSHN · OASIS test, evaluated at phi=0.5 | Dice | Improved |
| train SSHN with r = 0 (constant phi = 0.5) (ablation) | SSHN · OASIS test, phi=0.5 | Dice | Hurt |
| widen prior to U(0.4, 0.6) beyond narrow ranges (ablation) | SSHN · OASIS test | Dice | No clear effect |
| reduce rescaling factor from 0.5 to 0.4 | SSHN · OASIS and PanDental | inference FLOPs: over 50% reduction | Improved |
| reduce rescaling factor from 0.5 to 0.3 | SSHN · OASIS and PanDental | inference FLOPs: 70% reduction | Improved |
Closest papers by meaning
- EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks (ICML 2019 · similarity 0.86)
- Streaming Convolutional Neural Networks for End-to-End Learning With Multi-Megapixel Images (TPAMI 2020 · similarity 0.86)
- Scaling Recurrent Models via Orthogonal Approximations in Tensor Trains (ICCV 2019 · similarity 0.86)
- LVM-Med: Learning Large-Scale Self-Supervised Vision Models for Medical Imaging via Second-order Graph Matching (NeurIPS 2023 · similarity 0.86)
- LoMix: Learnable Weighted Multi-Scale Logits Mixing for Medical Image Segmentation (NeurIPS 2025 · similarity 0.84)
- The Effect of Intrinsic Dataset Properties on Generalization: Unraveling Learning Differences Between Natural and Medical Images (ICLR 2024 · 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.
