EfficientNet training for biological data

Has anybody tried EfficientNet ( https://arxiv.org/abs/1905.11946 ) ? It seems quite promising as the authors claim to reach accuracy comparable with the state of the art and even surpass it while having significantly less parameters. I am currently trying to implement and train it in keras using my dataset for virus infected/non-infected nuclei.

Hm, looks impressive! Thanks for the paper!

I had a look at their repo:

Unfortunately seems to be done in TensorFlow2 might need swing up a new env before I can try it.