Caffe: Efficient Net, GPipe or Amoeba Net implementation for Caffe?

[adriacabeza]

I was wondering if there is (or will be) any version of these classifications state-of-the-art architectures implemented in Caffe since the DL game changes so fast but the model zoo does not.

[TechnikEmpire]

@adriacabeza This repo has been in maintenance mode for a long time and you can consider it dead. Repo owners: change my mind.

[beniz]

Not exactly what you are looking for, but we (jolibrain) are still maintaining a widely enhanced fork here: https://github.com/jolibrain/caffe
It is missing some stuff from late additions to Caffe, but has many more layers, data augmentation and other useful stuff.
Model architectures can be found here: https://github.com/jolibrain/deepdetect/tree/master/templates/caffe
Not efficientnet, etc.. at the moment.

[strawberryfg]

Not what you are asking about, but it seems that this work is based on Caffe? https://arxiv.org/abs/1903.06586?

And the author's previous work SENet is also implemented in Caffe, which is great.

I've successfully transformed a PyTorch model to Caffe model, so I guess you'll have to transform the weights? There might be some bias in converting, like zero paddings. But that won't be a big issue, I suppose.

[beniz]

So EfficientNet for Caffe available here: https://github.com/jolibrain/deepdetect/pull/726