zyj061 / retina-crnn_model Goto Github PK

Hello, I am having trouble performing the training of the CNN model for natural scenes movie 2. As I do not have access to how the dataset with the frames and labels is preprocessed I think it might have to do with that.

I was wondering if you could kindly share a code snippet with how and in which way you extract the frames and labels from the dataset indicated in the paper (respectively video frames from movie 2 and neuronal recordings evoked when watching the same movie) or possibly the fully trained CRNN model with weights.

I would like to thank you in advance for your time.

Hi! I read the paper seriously about the RGC data but don't find the way to generate the dataset in the source code. So I am seeking help here.
The model is trained with the response of 80 RGCs. However, there are 38/49 RGCs found in different retina. Besises, there are 1800 frames and 1600 frames in movie1 and movie2 respectively. A single frame will display about 33 ms in front of the retina. But there are only responding neural activity of 60 scenes of length 300 ms offered in the response of RGC data. So I am confused about how the 60 sences are selected? The orginal paper gives the method but I can not find the final selected scenes?
So can you offer more details about how to generate the dataset used in your code or offer the dataset directly?

Thanks!