quality/resolution of image results

[SB2020-eye]

Hello again.

The images that result from running docExtractor (as found in the "text" and "illustration" folders) -- should I expect these to be full resolution with respect to the original? Or is some reduction or loss involved?

(I did my own little (if crude) test. Here is a clip resulting from docExtractor:

The file size is 33.7 KB.

Here is a crop I made from the original:

The file size is 70.4 KB.

I confess I don't know enough about digital imagery, how files are saved, etc to know if this concludes anything or not. :) Regardless, my goal is to have crops made using docExtractor that are lossless.)

[monniert]

Hi SB2020-eye, indeed I didn't provide much details about the quality of extracted elements, but I can understand lossless extractions are key when working with HD documents:

• the prediction is made at a resolution within 1280x1280 but is then upsampled to match original image dimensions, so the crops are made at full resolution wrt the original image,
• there may be some loss when saving the crop image to disk, especially when saving as a jpg file. There are 2 main options that you can try to avoid losing too much information: i) play with PIL save function arguments (line 96-97 in extractor.py), you can start by setting quality=95 and look at the results (see here for more information), ii) save images in a lossless format like png, you can do that by passing out_ext='png' when instanciating the Extractor (line 212, extractor.py)

Be aware it will take much more space than compressed jpg images, I hope this helps!

[SB2020-eye]

This is all very helpful, indeed! Thank you so much!