EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks RT-1-X uses EfficientNet + FiLM.