AI Paper Atlas

Research map / Image generation / Super-resolution

Learning Filter Basis for Convolutional Neural Network Compression

Yawei Li, Shuhang Gu, Luc Van Gool, Radu Timofte

ICCV 2019 · Image generation › Super-resolution · approach: CNN and transformer SR architectures · 105 citations (OpenAlex)

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
add basis sharing between the two convolutions in a residual block (m=16) (ablation)Basis-compressed EDSR · Set5 x4PSNR (dB): 32.14 -> 32.16No clear effect
add basis sharing within residual block (m=64) (ablation)Basis-compressed EDSR · Set5 x4PSNR (dB): 32.33 -> 32.30No clear effect
add basis sharing within residual block (parameters) (ablation)Basis-compressed EDSR (m=64) · EDSR one residual block#Params: 106k -> 70kImproved
increase the number of basis from 16 to 64 (ablation)Basis-compressed EDSR · Set5 x4PSNR (dB): 32.14 -> 32.33Improved
increase the number of basis from 24 to 32 (ablation)Basis-compressed DenseNet-12-40 · CIFAR-10top-1 error (%): 5.69% to 5.57%Improved
add compression of the 1x1 transition layers (6 splits) and use more basis (ablation)Basis-compressed DenseNet-12-40 (M32 -> M36T6) · CIFAR-10top-1 error (%): 5.57 -> 5.32Improved
increase transition-block splits to 12 with 38 basis (ablation)Basis-compressed DenseNet-12-40 (M36T6 -> M38T12) · CIFAR-10top-1 error (%): 5.32 -> 5.56Hurt
implement split-wise decomposition with s shared-weight 2D convolutions instead of a 3D convolutionCompressed EDSR · training iteration, batch 16, 48x48 patchesrun time (ms): 121 ms -> 62 msImproved
use basis learning instead of FactorVGG-16 compression · CIFAR-10top-1 error (%): 7.12 -> 6.18Improved
use basis learning instead of FactorEDSR compression · Urban100 x4PSNR (dB): 25.98 -> 26.36Improved
add basis sharing (Basis-S instead of Basis) (ablation)EDSR compression · Set5 x4PSNR (dB): 32.39 -> 32.29Mixed
use basis learning instead of Factor-SIC3SRResNet compression · Set5 x4PSNR (dB): 31.86 -> 31.90Improved

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