EfficientNet-V2 weights ported from Tensorflow impl

Weights from https://github.com/google/automl/tree/master/efficientnetv2

Paper: EfficientNetV2: Smaller Models and Faster Training - https://arxiv.org/abs/2104.00298