27 comments

[ 4.1 ms ] story [ 44.3 ms ] thread
what the hell, i tried a few images from unplash and it worked flawlessly. how does this magic work?
i've tried maybe 30 images now. All of them it nailed. i did find one that made me laugh. it doesnt know what a goat is.

http://picjumbo.com/wp-content/uploads/IMG_9454-1300x866.jpg

["dog", "australia", "deer", "bear", "cow", "mouflon", "lemur", "wild", "safari", "zoo"]

They must have an army of trolls who are typing in tags.
I tried a few images, for all of them "woman" was the first result (only one image had a woman in it).
Adam from Clarifai here.

As tommoor pointed out, this is just a thin wrapper around our demo at http://www.clarifai.com/

(we're very flattered...)

A real API will be out soon. It won't be throttled as heavily as the demo, and will be more developer friendly.

you can sign up for early beta access at clarifai.com.

We've changed the title (from "Identifying Images With an API that Actually Works") in an attempt to share the credit fairly.
Can you roughly share when the developer API will be ready?
Wow. The classifier is really impressive. Will it be possible to train your own classifier on your service? I have a lot of clothing items that it would be useful to classify. I tried building my own with opencv, but I haven't had too much luck so far.

I signed up for the api access - would be very interested in playing a bit more with this.

Thanks!

Training custom classifiers isn't in the roadmap for v1, but there will be a mechanism for providing feedback (suggesting new tags and marking errors), and we'll continue to improve our models based on that.

If you have a very large (100k+ images) well-labeled repository to train from, send us a note at info@clarifai.com, we'll tawk.

pretty impressive!
Wow! This is really neat, I tried to find images that I didn't think it could process, the results are interesting.

http://upload.wikimedia.org/wikipedia/commons/1/1e/Blown_up_...

["piranha", "fish", "food", "water", "gold", "dish", "crab", "kitchen", "glass", "silver"]

https://www.flippers.com/images/See-SHFA1_Caps&Mods-PCB.JPG

["panel", "retro", "wine", "background", "design", "old", "tool", "letter", "art", "robot"]

http://history.nasa.gov/alsj/a11/a11_lpi_trvrsmap.gif

["background", "metal", "water", "man", "wall", "old", "abstract", "paper", "hand", "paint"]

Combine this with a Markov Chain and you get a nice story teller.
Well this went better than I expected :) EDIT: context- I'm the one that made the wrapper for clarifai's API, loved the service since I heard of it, great to see people appreciate (from the number of API calls you guys have made so far) both the service and the small wrapper to the API!
Try typing in

A python stacktrace about JSON

wow this is really cool! how is this done?! what kind of sorcery is this?!