AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Scaling-Translation-Equivariant Networks with Decomposed Convolutional Filters

Wei Zhu, Qiang Qiu, Robert Calderbank, Guillermo Sapiro, Xiuyuan Cheng

JMLR 2022 · Architectures and efficiency › CNN design · approach: Steerable rotation-equivariant CNNs

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 changedWhereMeasuredEffect
use ScDCFNet instead of SESNScDCFNet · STL-10accuracy (%): 83.89 -> 84.90Improved
use ScDCFNet instead of SESN, 2K trainScDCFNet · SMNIST 28x28 no augmentationaccuracy (%): 95.75 -> 95.87Improved
use ScDCFNet instead of SESN, 2K trainScDCFNet · SFashion 56x56 no augmentationaccuracy (%): 82.34 -> 83.01Improved
use ScDCFNet instead of regular CNN, 2K trainScDCFNet · SFashion 28x28 no augmentationaccuracy (%): 79.88 -> 81.32Improved
add scale data augmentation to ScDCFNet, 2K trainScDCFNet · SMNIST 28x28accuracy (%): 95.87 -> 96.68Improved
reduce spatial basis from K=15 to K=6 (add more basis functions) (ablation)ScDCFNet · SMNIST 28x28 2Kaccuracy (%): 95.61 -> 95.87Improved
reduce spatial basis from K=15 to K=6 (add more basis functions) (ablation)ScDCFNet · SFashion 28x28 2Kaccuracy (%): 80.88 -> 81.32Improved
increase K from 12 to 15 (ablation)ScDCFNet · SMNIST 28x28 2Kaccuracy (%): 95.88 -> 95.87No clear effect
increase scale basis Kα from 1 to 3 at K=15 (ablation)ScDCFNet · SFashion 28x28 2Kaccuracy (%): 80.74 -> 81.32Improved
use one-sided-replicate-padding in scale instead of zero-paddingScDCFNet · equivariance error vs depthrelative L2 equivariance errorImproved
use smaller scale filter support IαScDCFNet · equivariance error vs depthrelative L2 equivariance errorImproved
reduce spatial basis K in auto-encoder (K=5,8,12,15) (ablation)ScDCFNet auto-encoder · SMNIST reconstructionrelative L2 reconstruction errorImproved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy