search

Element-Research / rnn

Recurrent Neural Network library for Torch7's nn
BSD 3-Clause "New" or "Revised" License
941 stars 313 forks source link

Inputting images into 3 states of RNN #276

Closed RaviTej310 closed 8 years ago

RaviTej310 commented 8 years ago

How can I build an RNN model with three unrolled states(t-3,t-2 and t-1) which accepts three images of dimensions 200x200 and predicts a fourth image? What would I need to change in this RNN example?

viorik commented 8 years ago

@RaviTej310 in case you want to use a convolutional LSTM, you could have a look here for an example of predicting the next frame in a sequence.

RaviTej310 commented 8 years ago

I would not like to use that link as it uses cunn and cutorch for which I need to install CUDA and everything. This is unfortunately not possible for me. Is there any way I can modify the code in the link I have given?

viorik commented 8 years ago

@RaviTej310 I am not sure what is needed to make that example work on images. The example in the link I sent can be run on CPU as well, if instead of using the SpatialConvolutionalNoBias from extracunn, you use the normal SpatialConvolution and call noBias to cancel the bias. I can advise you if you want to make that work.

RaviTej310 commented 8 years ago

@viorik Sure. Please advise. Thanks for the help! What do I need to change in the code? Shall I create an issue in your repo? I am kind of on a tight schedule.

viorik commented 8 years ago

I've uploaded here a simple demo of the CPU version. Please post there to let me know if you manage to run it or not. cheers.