Hello,
I have failed to reconstruct the moving mnist sequences.
At first, the loss became nan. Then, I increased the fac parameter to 0.7 and no longer encountered nan values, but it seems the shape module fails to converge. Temporal features are learned according to reconstruction images created during training, but shape features aren't.
I was wondering if you could provide the configuration parameters you used for both datasets?
Hello,
I have failed to reconstruct the moving mnist sequences.
At first, the loss became nan. Then, I increased the fac parameter to 0.7 and no longer encountered nan values, but it seems the shape module fails to converge. Temporal features are learned according to reconstruction images created during training, but shape features aren't.
I was wondering if you could provide the configuration parameters you used for both datasets?