Option to only download samples / thumbnails?

[can1357]

Although I've searched all the options in Grabber I wasn't able to find a "download thumbnails" or "download samples" option. I am a deep-learning researcher and the ability to download low-quality alternatives is very convenient when batch downloading > 20k images.

It would be awesome if this could be supported officially.

Excuse me if this feature already exits and I was incapable of finding it, Can

Temporary solution for windows users: Byte-patch [Grabber.exe + 0xD7507] to 0xC (Build: 25 May)

[Zekken01]

If I may ask, What are you trying to teach an AI with images downloaded from boorus?

[can1357]

@brodycas3 auto tagging pictures & feature extraction. So far in my experiments I was able to do searches in untagged pictures (not from boorus) based on the training done using tagged pictures (from boorus) similar to: get_vector("blond") + get_vector("fang") or get_vector(picture) - get_vector("naked"). Quite fun actually.

[Zekken01]

That seems pretty interesting.

[MasterPetrik]

@can1357 maybe you will be interested about already existed tag logging feature https://github.com/Bionus/imgbrd-grabber/issues/874

and do you heard about Hidrus project?(Sorry, Bionus) It's anime DB with cloud tags storage, marking by community. There are exist almost 15 GB hash-tags info(downloads separately) about almost all images from all boorus https://github.com/hydrusnetwork/hydrus and here is a thread with all tag archives for it https://8ch.net/hydrus/res/2651.html

And moreover, there is already exist anime images tag predicting deep AI neetwork: http://demo.illustration2vec.net/# And it works almost perfect:

So if you gonna work with tag recognition, I hope some of this will help you, good luck!

I'm interested in mashine learning by myself(as user), so If you will have some progress, please tell me about that)

[can1357]

@MasterPetrik I didnt know about the tag logging feature nor about the Hidrus project; good to know about! I'll definitely let you know if I have some notable results. Illust2Vec seems to work quite good though; quite impressive.

[MasterPetrik]

looks like they even giving they Pre-trained model for free: http://illustration2vec.net

[Bionus]

There is no current way to download thumbnails. There is an option in the settings to download samples but I can't guarantee you it's working properly on all source.

I could add a new CLI option to download or return the URLs of thumbnail/samples though. Depending on the sources you're using, you could also directly build these URLs from the full image URL.

If the source is danbooru-based for example, the image https://danbooru.donmai.us/data/93e6d0d665c373bd6c5c9df8088b501e.jpg will have a thumb located at https://danbooru.donmai.us/data/preview/93e6d0d665c373bd6c5c9df8088b501e.jpg: simply replace /data/ with /data/preview/. Same thing for samples, except you have to replace /data/ with /data/sample/sample-, and not all images have samples.

[can1357]

The bytepatch I do changes: https://github.com/Bionus/imgbrd-grabber/blob/88bfbd66b2693387f05bf0f33d2bb7ad8737dd2a/lib/src/models/image.cpp#L203 m_url = m_fileUrl.toString(); -> m_url = m_previewUrl.toString(); and it seems to work fine for most images. It'd be really cool if you could add it @Bionus.

Where's the download samples option thought I cant seem to be able to find it 😕

[Bionus]

The bytepatch I do changes

You also have m_sampleUrl if I remember correctly.

It'd be really cool if you could add it @Bionus.

I'll add it as a CLI option in the next update. Since it seems you're building from source (:question:) you can grab the develop branch when you see one of my commits referenced in this issue.

Where's the download samples option thought I cant seem to be able to find it 😕

If you don't check the first box ("Download original images") it should download samples when available.

[can1357]

I was byte-patching the already-built executable file because I do not like to deal with Qt. Anyway, thanks a lot for considering the addition of this feature :)