How to model Recurrent Convolutional Neural Networks in Caffe?

Question

I am quite new to the Caffe framework, only recently starting to use it.

I understand that modelling CNNs is allowed, however, is it possible to combine RNNs (not much experience with these) and CNNs together, to form a recurrent convolutional neural network?

I am currently looking at: http://caffe.berkeleyvision.org/tutorial/ for Caffe knowledge and following some tutorials online. Any other tutorials which you recommend would be appreciated.

Thanks.

For what reason do you want to do this? – Bastian Dec 15 '17 at 08:32 — Bastian, Dec 15 '17 at 08:32

score 1 · Answer 1 · answered Dec 26 '17 at 20:09

1

Although Caffe provides Recurrent Layer, unfortunately the documentation is terrible. This is the only legit example of LSTMs that I know of that is written in Caffe: https://people.eecs.berkeley.edu/~lisa_anne/LRCN_video

answered Dec 26 '17 at 20:09

dusa

840
3
14
31

score -1 · Answer 2 · answered Dec 14 '17 at 22:40

-1

Look at caffe'a "Recurrent" layer, there's also an "LSTM" layer.
If you are looking for something more complex, you might consider unrolling the net yourself, see this answer for an example.

answered Dec 14 '17 at 22:40

Shai

111,146
38
238
371

I considered both but I dont think any of them is convolutional. I think unrolling is the only way to implement it. – user3126802 Dec 14 '17 at 22:46

How to model Recurrent Convolutional Neural Networks in Caffe?

2 Answers2