Subscribe
Sign in
CLIP API by Booste — Use OpenAI's newest image classifier with one line of code.
Booste APIs boiling all the pains of the production ML stack into a single line of code.
We're super excited to launch an API to OpenAI's powerful new no-training-needed image classifier to Python and Node. Give it a spin!
20
Replies
This is exciting stuff @erikdoingthings! I've been wanting to play with ML for a while but always been discouraged by how much setup seemed to be necessary to even get from 0 ->1 in terms of "productizing". Might have to give old ideas another try now that booste is making it actually possible! Congrats on the launch! 🎉
https://paint.wtf is so much fun! This API will make ML-oriented products much more approachable, especially as more and more state of the art, open source models hit the airwaves.
Great work!
What's up Product Hunt!
Second time ever launching here :)
I'm Erik, the maker behind Booste. My goal is to take the entire production Machine Learning stack and boil it down to just one line of code, like Stripe did to payments and Twilio did to communications. ML-as-an-API
I'm a self-taught dev myself, and love abstractions. When I dove into the MLOps world, I found a dumpsterfire... I'm going through the pains of productionizing the newest open-source models, so you don't have to.
About a month ago, OpenAI released an incredibly interesting model called CLIP
https://openai.com/blog/clip/
What's special about it is that it's one of the first "Zero-shot" image classifiers, meaning it can predict a connection between a class (IE: "a dog") and an image, without ever being explicitly trained on that class.
The use-cases are endless:
- Advanced image search
- Video indexing
- https://paint.wtf, an AI-judged MS paint competition
- Hotdog / Not Hotdog app
- Finally seeing that the dress is Black/Blue, not Gold/White
I've thrown it up onto Booste for y'all to play with - free for the month of February!
You can install into your python/node environments using the product link, or just run it straight away in this Colab notebook:
https://colab.research.google.co...
You can always reach me at erik (at) booste.io :) I want to hear about the cool stuff you build.
Important meme attributions:
I officially credit to @casey_caruso and @lstephanian for the killer "The Dress" meme used here. I'm in camp Gold/White, so was emotionally crushed when they first ran that test.
The Nickelback GIF, on the other hand, was my own meme genius at work and I take all credit.
Congrats on launching, Erik!! I'm a huge fan of image classifiers so this is definitely something I'll be spending a lot of time with! Thanks for building it! 🤩
@steventey Thanks Steven!! Love having an ML person like yourself play with it! LMK if you have any thoughts or needs on it.
I built a garbage vs recycling classifier last year and by far 90% of the work was spent productionizing our model for the app (making it scalable, ensuring the frontend could call backend without crashing etc). Having an out-of-box reliable API for our use case would be incredible.
Excited to see where this is going.
@morriscode Would to link up and talk about how you did it! IE onprem/cloud, Sagemaker/self, etc.
@erikdoingthings Sure, you can see the code @ https://github.com/kylejmorris/d... From what I remember, we wanted to use an inception model on a phone but it was too slow even on my laptop, so we tried SqueezeNet instead. Realized it's just going to be a pain to port the working code from laptop to phone, so we tried getting a flask server setup but bailed on it since we didn't even have the model working well yet so why productionize it (probably should have taken this route further in retrospect). Ended up just using google colab to get something roughly working, never got it on phone. Such is the tragic ending to this ML saga.
This product is definitely going to help me become more confident with ML, and it will give me the opportunity to get creative! Thanks Erik
Stack Roboflow