error terminate called after throwing an instance of 'std::bad_alloc' about keras-ocr HOT 5 CLOSED

Hello! Could you provide a few more details to help me reproduce?

• TensorFlow version (import tensorflow as tf; tf.__version__)
• Keras version (import keras; keras.__version__)
• Did you try to run the demo code more than once with the same result?
• Was any other code executed before the demo code?

from keras-ocr.

Hi faustomorales, some detail of my system is:

TensorFlow version: 1.14.0.
Keras version: 2.3.1

"Did you try to run the demo code more than once with the same result" yes, I run it more than 5 times, and totally crash.

"Was any other code executed before the demo code?" No, I just create a new test.py file and paste your example code.

It is show error after run this line "boxes = detector.detect(images=[image])[0]", the memory increase very fast (from 4GB to 11GB in 2-3 second), and crash.
log in terminal:

terminate called after throwing an instance of 'std::bad_alloc'
what(): std::bad_alloc
Aborted

Thank you.

from keras-ocr.

I ran the demo code with that combination and was unable to reproduce. Some more things to check --

1. What version of OpenCV do you have (import cv2; cv2.__version__)
2. Could you verify that image.shape is (480, 640, 3)? I'm wondering if the package is reading the file incorrectly somehow.
3. You could try running the demo code with TensorFlow 2.0.0 by doing pip uninstall -y tensorflow then pip install tensorflow==2.0.0.

Thank you for your patience!

from keras-ocr.

Hi faustomorales, thank you for fast reply,

my image size is (3024, 4032, 3). I forget to resize it into (480, 640, 3).

When i resize to (480, 640, 3), it work.

Thank you very much.

from keras-ocr.

Glad that worked out for you! Just to clarify, you don't have to use (480, 640) exactly. The model should still work for different dimensions / aspect ratios.

from keras-ocr.