I was thinking about putting the 128x128 frames back together before calculating the losses.
In the file Teco.py, the loss is calculated only on the single 128x128 patch of the generated image.
I assume batch = 4 means we are dealing with 4 different patches of different videos frames.
And RNN_N being equal to 10 means we have a sequence of 10 patches from 10 consequent frames?
So I was wondering if we can somehow obtain the output of all the 128x128 patches of a frame or not.
I was thinking about putting the 128x128 frames back together before calculating the losses. In the file Teco.py, the loss is calculated only on the single 128x128 patch of the generated image.
For example, this https://github.com/tom-doerr/TecoGAN-Docker/blob/e78b1d2f3d3e9ae00df7b5a123118f375167a712/lib/Teco.py#L168 is the s_gen_output that is afterwartds used in https://github.com/tom-doerr/TecoGAN-Docker/blob/e78b1d2f3d3e9ae00df7b5a123118f375167a712/lib/Teco.py#L320 to compute the loss. Is there a way to reach out all the patches of a single frame or not?
I assume batch = 4 means we are dealing with 4 different patches of different videos frames. And RNN_N being equal to 10 means we have a sequence of 10 patches from 10 consequent frames? So I was wondering if we can somehow obtain the output of all the 128x128 patches of a frame or not.