WideResNet architecture
Best resnet for Cifar10/Cifar100
https://arxiv.org/abs/1605.07146
You could take a look at this with SAM or in the resnet family at @leondgarse https://github.com/leondgarse/keras_cv_attention_models#resmlp
See also: https://paperswithcode.com/sota/image-classification-on-cifar-100
Would like to work on this.
Hi @LukeWood ,I would love to implement WideResNet duing GSoC. I'm currently drafting my proposal and would love to know if I could implement this during GSoC ?
Any update on this feature? cc @blackhat-coder
Hi @innat, This feature proposal wasn't accepted, however I can still work on implementing this architecture.
Do you mean, didn't accept in GSoC? If so, I don't think it's required to pass things in GSoC in order to be eligible to include in KerasCV. ( WideResNet is widely used with over 5288 citations. )
Can you assign this to me as well? :)
Thanks!
Small update, got to this last night, and I'm almost done (I think). Just gotta tweak a few things and test it thoroughly on different setups (they have a few listed in the paper)
Seems to be done, just waiting for the training to finish to validate whether it behaves the same as in the paper. Sending the pull request in a few hours!
Lets deprioritize this until theres a strong use case; and focus on migrating the existing backbones!
@LukeWood this network is cited by 6776 times, here is the pwc report.