How do i validate a model on new data? can you provide stepwise solution to transform image and predict it on new data?

[jonomon]

You could look at this notebook to get an idea of the entire pipeline.

If you want to validate the model on new data, simply edit the "Dataset creation" section in the notebook with your new data.

[shaheerakr]

@jonomon yeah thats cool but how do i provide my own data in this section? because it downloads data from: http://www.fki.inf.unibe.ch and pre process it, i want to provide my own images to be pre processed.

[Asus1996]

Hi can u provide step to teach us how to use our own data~ how to edit the "Dataset creation' section ?

[shaheerakr]

Here is the link to my repository: https://github.com/shaheerakr/cheating-detector you need to get the ocr folder (it is modified a bit for this purpose) from this and checkout the ocr.ipynb for step wise procedure of converting images to texts. or you can checkout image_to_text.ipynb if you want to create a REST API using flask to convert from image to text. Leaving this issue open until a proper solution is here, Thanks

[ouardafenek]

Dear @jonomon, thank you so much for the work you have done. It helps a lot. Though I'm having a little problem to test the algorithm on images of my own. I'm using the 0_handwriting_ucr notebook and as @shaheerakr said above i strugle replacing the dataset creation part. I've tried tried replacing the 4 images that you sampled with images not from the IAM dataset but I'm having shapes error later. Thanks in advance.

[jonomon]

@ouardafenek you can use the resize_image function to resize your input image.

[Pabper]

@jonomon Do other images need to mantain the aspect ratio of the ones in the IAM Database?

[jonomon]

@Pabper That would ideal however, the model should have some tolerance as the writers have different writing styles.