ImageNet-trained CNNs are biased towards texture; increasing shape bias improves accuracy and robustness

ICLR 2019 Robert GeirhosPatricia RubischClaudio MichaelisMatthias BethgeFelix A. WichmannWieland Brendel

Convolutional Neural Networks (CNNs) are commonly thought to recognise objects by learning increasingly complex representations of object shapes. Some recent studies suggest a more important role of image textures... (read more)

PDF Abstract

Results from the Paper


Ranked #3 on Domain Generalization on ImageNet-R (using extra training data)

     Get a GitHub badge
TASK DATASET MODEL METRIC NAME METRIC VALUE GLOBAL RANK USES EXTRA
TRAINING DATA
RESULT BENCHMARK
Domain Generalization ImageNet-A Stylized ImageNet (ResNet-50) Top-1 accuracy % 2.3 # 6
Domain Generalization ImageNet-C Stylized ImageNet (ResNet-50) mean Corruption Error (mCE) 69.3 # 3
Domain Generalization ImageNet-R Stylized ImageNet (ResNet-50) Top-1 Error Rate 58.5 # 3

Methods used in the Paper