EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks This paper addresses the challenge of scaling Convolutional Neural Networks (ConvNets) to achieve better accuracy and efficiency. The authors propose a compound scaling method that uniformly scales network depth, width, and resolution using a set of fixed coefficients, which they determine through a small grid search. This method is demonstrated to be effective on MobileNets and ResNet, leading to improved performance. Further, they use neural architecture search to design a new baseline network and scale it up to create a family of models called EfficientNets. EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller and 6.1x faster than the best existing ConvNet. EfficientNets also perform well on transfer learning datasets, achieving state-of-the-art accuracy on five out of eight datasets with significantly fewer parameters. The paper provides a comprehensive study of ConvNet scaling and introduces a principled approach to model scaling, demonstrating its effectiveness through extensive experiments.EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks This paper addresses the challenge of scaling Convolutional Neural Networks (ConvNets) to achieve better accuracy and efficiency. The authors propose a compound scaling method that uniformly scales network depth, width, and resolution using a set of fixed coefficients, which they determine through a small grid search. This method is demonstrated to be effective on MobileNets and ResNet, leading to improved performance. Further, they use neural architecture search to design a new baseline network and scale it up to create a family of models called EfficientNets. EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller and 6.1x faster than the best existing ConvNet. EfficientNets also perform well on transfer learning datasets, achieving state-of-the-art accuracy on five out of eight datasets with significantly fewer parameters. The paper provides a comprehensive study of ConvNet scaling and introduces a principled approach to model scaling, demonstrating its effectiveness through extensive experiments.