search

NHZlX / papers

1 stars 0 forks source link

NasNet: Learning Transferable Architectures for Scalable Image Recognition #4

Open NHZlX opened 6 years ago

NHZlX commented 6 years ago

https://arxiv.org/abs/1707.07012 Submitted on 21 Jul 2017

NHZlX commented 6 years ago
1bdb7ad4cc2f1ea7353fdad8cefc3dae 0f7399c1e04e3a3089e667a030929447 18419310495a174dfa64daca19e6cc84
NHZlX commented 6 years ago

I am more concerned about the model of the mobile. NasNet-A(4 @ 1056), 4 @ 1056 represents the value N and output channel of the last second layer in Figure 2.

I download the NasNet-A(4 @ 1056) model, the model size is about 50Mb(gzip), and five times bigger than mobilenet. The Flops is about 564M, same with the mobilenet.

The speed comparison needs to be tested.