ValueError: operands could not be broadcast together with shapes about keras-ocr HOT 3 CLOSED

The API for recognize was changed to support multiple images as of #9. In this case, if you change pipeline.recognize(image) to pipeline.recognize([image])[0], you'll be off to the races. I'm sorry for the breaking change -- I think the API is rapidly stabilizing thanks to the great feedback provided by yourself and others.

I made this change because going from one image to multiple images efficiently (i.e., with appropriate batching at each step) is non-trivial for users, so I wanted to make the API handle that for them. The examples all assume multiple images because doing so requires no additional complexity in the example code and is easily adapted to a single image, as the fix above shows.

With regard to examples, I agree with you that it's a good idea to include a detector-only example and a recognizer-only example along with the existing full pipeline example. That said, the examples should all use multiple images, again because it's easy for users to adapt a multi-image example to a single image but it is not as easy to conceptualize doing the reverse. Would you want to send a PR with the added examples in /docs/examples/using_pretrained_models.rst?

from keras-ocr.

I believe the above answer resolves the error you're seeing so I'm closing this issue for now. If it does not resolve your error, please reply here and I will reopen. Thanks!

from keras-ocr.

Thank you very much for your fast response.
I tried to understand whats happening and it's quite clear now. Because of the change to support multiple images I had to pass a list of images (pipeline.recognize(image) to pipeline.recognize([image])) and since the output is a list of lists I had to extract the list with [0] before passing it to keras_ocr.tools.drawBoxes ...

I would very much like to provide examples for the API but never really worked with git and PRs and such :) I added this task to my todo list but even if its not a big deal I will need some time to get to this. Till now I just played around when I got some free time (weekend) to see whats possible.

from keras-ocr.