Christian Bartz

Yes, this is actually intended :sweat_smile: Our original work came from the idea that we want to extract one box per word with multiple characters. However, we thought what if...

Hmm, it seems to me that the localization network never felt the need to converge to localize individual characters as the task for the recognition network was too simple. You...

You could also try to lower the learning rate of the recognition network to encourage the localization network to try harder to make t easier for the recognition network.

Yes, it might not be necessary to use the affine matrices. You could also just train the recognition network on patches you extracted from a regular sliding window. So basically...

Nice, that's good to hear. And the image looks the way it is supposed to :+1:

No, you can not predict the content of multiple words per image without retraining. The code could be used to do exactly that, but the text recognition model provided by...

Hi, we did some experiments with different chunk sizes (as reported in the paper) and found that having a chunk size of 10 seconds works better than a smaller chunk...

Hmm, difficult to tell what's wrong. What did you do exactly?

I also don't know what's wrong. I can only advise you to do some debugging and determine why the array `probabilities` seems to have only one axis instead of the...

Seems, that your model does not return any predictions. You could check that your are using the correct model and that your model gets some input at all.