Need help getting this to work.

[arthurwolf]

Hello.

With a lot of work, I got demo.py to work, but now I'm stuck.

I can get it to do panel cutting/detection, but I have no idea how to get it to do bubble detection.

I read all the documentation, and some of the source code, and I'm really not clear on how to get this to work.

Any help would be extremely welcome.

@hanish3464

Thank you!!

[arthurwolf]

I partially got it to work, it just was too strict on bubble detection so I wasn't seeing any detected, until I saw one at random after a few weeks, and figured out I could see more by adjusting --cls.

I would like help/guidance on how to improve detection of both panels and bubbles.

There is an option for large image where it seems to cut images, but no real guidance on how to use it. What is it meant to do/be used for, and how should I configure/use it?

If I let it cut images in multiple parts, wouldn't it then be unable to detect panels/bubbles that «overlap» two images ?

Also, I see in the large image mode it resizes images to 720 width. Is there a reason for this, like maybe this is an optimal width and I should always resize all images to that width before feeding them to the model (maybe because the model was trained at that size?)

Also, is there some way I could train a new model with my own dataset? (I work more with comics/manga than webcomics/webmanga)

Thanks for any assistance anyone is able to provide. Also if you know any way to contact the author of the project I would really appreciate it. My email is wolf.arthur@gmail.com.

Best regards.

[arthurwolf]

I'm looking for help getting this code to run with the latest pytorch (or migrate it to tensorflow etc).

I'm ready to pay (a bit) for help with this.

If anyone else is interrested in this, I'd also love to chat about it.

[mwmw7]

Hello I am Branden Let's do it together