Architectures

We collect 40 state-of-the-art Deep neural network architectures already trained on ImageNet-1k.

Performance indices

We consider multiple performance indices useful for a comprehensive benchmark of DNN models. Specifically, we measure:

• Accuracy on ImageNet
• Model complexity
• Memory usage
• Inference time
• Accuracy vs. Computational complexity