search

yajiemiao / pdnn

PDNN: A Python Toolkit for Deep Learning. http://www.cs.cmu.edu/~ymiao/pdnntk.html
Apache License 2.0
224 stars 105 forks source link

additional input #4

Closed tofigh- closed 9 years ago

tofigh- commented 9 years ago

Hi, Assume a Conv network consisting of several conv layers and several FC layers. Is it possible to connect some of the inputs directly to a FC layer ,i.e., for some inputs can we skip the Conv layers?

ghost commented 9 years ago

It's possible. In models/cnn.py, the outputs of the i-th layer are represented by the tensor variable self.lalyers[i].output. You can send this variable as the input to upper layers.

I guess you are talking about the multi-stage style feature fusion http://arxiv.org/pdf/1204.3968.pdf Then, inputs from multiple lower layers can be combined with theano.tensor.concatenate. Refer to https://github.com/yajiemiao/pdnn/issues/2