Machine Learning Scanlations

Joined
Apr 23, 2018
Messages
1,071
This is unreadable and you absolutely have to get a proofreader until MTL becomes more tolerable. However, I would love it if you became an example of what MTL is capable of.

Please utilize 'Content Aware' algorithms such as Content aware fill for cleaning. This has the advantage of being capable of doing passable level of redrawing on simple backgrounds. Solid backgrounds are easy enough to detect to turn it off for them. More complex backgrounds like patterns and faces may or may not be passable, depending on the algorithm in use.

Just do what you are currently doing, except instead of sending the OCR detected text-boxes into a solid box-fill tool, send it to a content aware fill tool. If you don't already, you may have to first turn the box into a selection, then run a content aware fill filter instead (steps needed depends on program used).
 
Aggregator gang
Joined
Mar 5, 2019
Messages
774
If you are not trolling, I 100% recommend you get some, what is considered good, fan translations, then get the raw and feed it in, and have the mla check against known good till it figures it out.

This SHOULD be able to get a good end product, from there, have someone who can type the stuff out in japanese input it and that should do it. from there time is the resource most needed.
 

Users who are viewing this thread

Top