Fixing the train-test resolution discrepancy

Hugo TouvronAndrea VedaldiMatthijs DouzeHervé Jégou

   Papers with code   Abstract  PDF

Data-augmentation is key to the training of neural networks for image classification. This paper first shows that existing augmentations induce a significant discrepancy between the typical size of the objects seen by the classifier at train and test time... (read more)

Benchmarked Models

RANK
MODEL
REPO
CODE RESULT
PAPER RESULT
ε-REPRODUCED
BUILD
1
FixResNeXt-101 32x48d
86.4%
86.4%
2
FixPNASNet-5
83.7%
83.7%
3
FixResNet-50 CutMix
79.8%
79.8%
4
FixResNet-50
79.1%
79.1%