Hello Sup Forums. I'm training a DNN to identify smug anime images and I need a larger database of them in order to get better accuracy. ITT can you dump images that should be in the dataset. It doesn't just need to be smug anime. The classifier needs to differentiate between anime and non-anime as well as smug and not smug so ideally a mix of anything would do as long as there are enough smug anime girls.
Also Discuss AI.
You need to first define what counts as smug first.
...
Found the autism.
if that was well defined then there would be no need for op to teach a neural net.
it's an "I know it when I see it" problem.
Just write a web scraper dumbass, do you really think you're going to get enough training data with a Sup Forums thread?
sounds like a fun project though. Which framework are you using?
tensorflow and I am using a web scraper to download threads worth of images but tagging them is really a pain in the ass. It's better to have a thread dedicated to smug anime mostly and then mark everything as smug anime unless otherwise noted. Faster training data.
Just search in google image "smug anime" and there you have it
If I were doing this I would just scrape the first n results of google image search 'smug anime face' or use the bing search api to get images for each category.. saves having to hand label
This is a shit thread
...
...
...
...
...
I will only continue if you promise to make your software and the trained net 100% free software
...
...
...
...
...
...
Smug Anime images are useful.
...
...
...
No. pic related
Get out
First of all, that's not anime OP. That's Miss Fortune, a character from the fighting game, "Skullgirls."
You could at least be right when you post something.
Google = botnet
Ok. Better?
That is my favorite smug anime girl image. You can't get smuger than that.
anime
Let's ask Yotsuba.
Is Sup Forums an anime image board?
...
How does one manage to become this smug?
If your DNN can get this one correct, it's doing pretty well