EfficientNetV2: Smaller Models and Faster Training | Google Brain Paper Explained
The newest version of EfficientNet v2 achieved better results on ImageNet top-1 accuracy than recently published NFNets, Vision Transformers, etc. This video walks through the new paper authored by Mingxing Tan and Quoc V. Le and explains 1) what is a progressive training and 2) Fused-MBConv layer and novel reward function for NAS. Paper Abstract: This paper introduces EfficientNetV2, a new family of convolutional networks that have faster training speed and better parameter efficiency than previous models. To develop this family of models, we use a combination of training-aware neural architecture search and scaling, to jointly optimize training speed and parameter efficiency. The models were searched from the search space enriched with new ops such as Fused-MBConv. Our experiments show that EfficientNetV2 models train much faster than state-of-the-art models while being up to 6.8x smaller. Our training can be further sped up by progressively increasing the image size during training, but it often causes a drop in accuracy. To compensate for this accuracy drop, we propose to adaptively adjust regularization (e.g., dropout and data augmentation) as well, such that we can achieve both fast training and good accuracy. With progressive learning, our EfficientNetV2 significantly outperforms previous models on ImageNet and CIFAR/Cars/Flowers datasets. By pretraining on the same ImageNet21k, our EfficientNetV2 achieves 87.3% top-1 accuracy on ImageNet ILSVRC2012, outperforming the recent ViT by 2.0% accuracy while training 5x-11x faster using the same computing resources.

00:00​ High-level overview 01:40​ NAS review 07:20​ Deep dive 15:35​ Novel reward 17:34​ Progressive training 21:10​ Stochastic depth regularization 23:00​ Results