ResNet

[BZbyr]

Using ResNet?

[KichangKim]

Yes. It is based on Microsoft's resnet implementation. https://github.com/microsoft/CNTK/tree/master/Examples/Image/Classification/ResNet

[DonaldTsang]

There are so many models to choose from

• https://github.com/BertMoons/Comparing-CNN-Architectures
• https://github.com/CeLuigi/models-comparison.pytorch
• https://github.com/yeephycho/nasnet-tensorflow

@BZbyr perhaps you can pick some examples that you would like to see.

[DonaldTsang]

@KichangKim if I am allowed to say this, are there any benchmarks or tests on multiple CNN architectures?

[KichangKim]

@KichangKim if I am allowed to say this, are there any benchmarks or tests on multiple CNN architectures?

In very early version of DeepDanbooru, I tested some types of CNN, like Inception, Squeeze-and-Excitation and so on. But not fully benchmarked it and chose ResNet for its simplicity and good reference code (in CNTK's repository). If you want, you can test yourself by extending deepdanbooru.model namespace.