Description
FixRes is an image scaling strategy that seeks to optimize classifier performance. It is motivated by the observation that data augmentations induce a significant discrepancy between the size of the objects seen by the classifier at train and test time: in fact, a lower train resolution improves the classification at test time! FixRes is a simple strategy to optimize the classifier performance, that employs different train and test resolutions. The calibrations are: (a) calibrating the object sizes by adjusting the crop size and (b) adjusting statistics before spatial pooling.
Papers Using This Method
DeiT III: Revenge of the ViT2022-04-14Three things everyone should know about Vision Transformers2022-03-18Training data-efficient image transformers & distillation through attention2020-12-23An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale2020-10-22Circumventing Outliers of AutoAugment with Knowledge Distillation2020-03-25Fixing the train-test resolution discrepancy: FixEfficientNet2020-03-18MaxUp: A Simple Way to Improve Generalization of Neural Network Training2020-02-20Big Transfer (BiT): General Visual Representation Learning2019-12-24Self-training with Noisy Student improves ImageNet classification2019-11-11Fixing the train-test resolution discrepancy2019-06-14