Inference with own image

[hasslercastro]

How do I transform my own images in a tf_record either to run on training or inference, (OCR, then stablishing vertext, etc ) ? Thank you, this is a nice project.

[TheClassyPenguin]

@hasslercastro have you found a way to do it?

I've been trying to build the pipeline myself digging a little through the code with a little help from the paper since it is a little hard to understand.

From my understanding this is what you need from an OCR/word area to fill in vertex_features:

• The upper left corner (x,y)
• Bottom right corner (x,y)
• Character length of the word

Keep in mind these are generated directly from the HTML and they are not from an OCR/CNN using the paper as the paper would lead you to believe.

Hope this helps a little bit towards building our own pipeline for custom examples.