AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Why do deep convolutional networks generalize so poorly to small image transformations?

Aharon Azulay, Yair Weiss

JMLR 2019 · Architectures and efficiency › CNN design · approach: Compact CNN architectures · 508 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
embed image at a size far from the training size (about 100 instead of about 220) (ablation)VGG16, ResNet50, InceptionResNetV2 · ImageNet embedding protocolP(top-1 change): approximately 10% -> approximately 30%Hurt
use an atypical image with low similarity to training neighbours (ablation)ImageNet CNNs · ImageNet crop protocolP(top-1 change): approaches 0.3Hurt
add bin-5 antialiasing (blur before stride)modern CNNs · ImageNet embedding protocolP(top-1 change): about 15%No clear effect
add bin-5 antialiasing, atypical imagesmodern CNNs · ImageNet crop protocolP(top-1 change): up to 20%No clear effect
train a readout from deeper layers instead of early layers (ablation)pretrained CNNs · ImageNet crop protocolP(top-1 change): below 5% -> higher with depthHurt
remove subsampling from the CNN (ablation)AlexNet-style CNN · CIFAR-10translation invariance: accuracies of about 0.8Improved
add extra augmentation of the same protocol as the testCNN · CIFAR-10translation invarianceMixed
use black fill instead of crop protocolVGG16, ResNet50, DenseNet121 (PyTorch) · ImageNet 1-pixel shiftP(top-1 change): 0.062 -> 0.18 (VGG16)Hurt
add ReLU (bias=5) after blur and subsamplingtoy 1D signal · Table 2 toy exampleglobal sum: 8 and 8 -> 0 and 6Hurt

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