AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Design of Kernels in Convolutional Neural Networks for Image Classification

Zhun Sun, Mete Ozay, Takayuki Okatani

ECCV 2016 · Architectures and efficiency › CNN design · approach: Filter design and trained-filter analysis · 20 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
use quasi-hexagonal kernels instead of 3x3 squareBASE (ImageNet CNN) · ILSVRC-2012 valtop-1 error (%): 31.2 -> 29.2Improved
use quasi-hexagonal kernels instead of 3x3 squareBASE (ImageNet CNN) · ILSVRC-2012 valtop-5 error (%): 12.3 -> 11.1Improved
use quasi-hexagonal kernels instead of 3x3 square (fewer params)BASE-A · CIFAR-10test error (%): 9.02 -> 9.10No clear effect
use random zeroed weights in 3x3 kernels (fragmented kernel) instead of quasi-hexagonal (ablation)QH-A · CIFAR-10test error (%): 9.10 -> 9.89Hurt
use random kernel shape patterns per layer instead of one fixed pattern (ablation)QH-A · CIFAR-10test error (%): 9.40 -> 9.10Improved
add extra dropout before global average pooling (ablation)BASE-A · CIFAR-10test error (%): 9.02 -> 8.71Improved
add extra dropout before global average pooling (ablation)QH-A · CIFAR-10test error (%): 9.10 -> 8.79Improved
use larger max-pool kernel (pool4) (ablation)QH-A · CIFAR-10test error (%): 9.10 -> 9.00Improved
use larger max-pool kernel (pool4) (ablation)BASE-A · CIFAR-10test error (%): 9.02 -> 8.87Improved
widen QH network to match BASE-A parameter count (QH-B) and larger QH-CBASE-A-AD · CIFAR-100test error (%): 31.2 -> 29.77Improved
use QH kernels with more feature maps (QH-C-AD)BASE-A-AD · CIFAR-10test error (%): 8.71 -> 8.42Improved
use QH-B-AD instead of base when only 1K training samplesBASE-AD · CIFAR-10 (1K training samples)test error (%): 44.9 -> 41.8Improved

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