About custom layers in DEL

[GaengSu]

(Before writing, I apologize if it is rude to you to randomly convert what you make.)

Hello :)

I'm so insterested in your network.

These days, Caffe framework doesn't update anymore, so I'm going to convert your network to Keras.

It's sloppy, but the conversion is almost over.

However, the problem comes up.

According to your 'train.prototxt', last layer('superpixel_pooling', 'sim_loss'sim_loss) use 'SuperpixelPooling', 'SimilarityLoss' type.

I think that type is made yourself.

That's why I can't convert those layers to keras. (The problem)

So, I need your help.. 🙌

I've already opened the "cpp" file, but I'm so stuck.

If you know how to convert easily your cpp file to something or how to use your network as a maintenance-enabled framework, tell me some hint plz..

Have a good day.

[yun-liu]

I feel sorry about that. I think the only way would be to understand the Caffe code first.

[GaengSu]

I feel sorry about that. I think the only way would be to understand the Caffe code first.

Thank you for your reply. 😥