What do you think of Clarifai? Have you used the API? Is it good for quick prototyping applications?
I’ve used Clarifai(and a bunch of other ML APIs) and I always found them to be little too overpriced, a little too generic and a little too slow on the inference part.
Whatever Clarifai and a tonne of other ML APIs can do, you can do too after the first 3-4 lessons of the class. The datasets they use are out in the open, the models they use are out in the open. Infact, a few days ago I discovered that the dataset(for the most part) that Cloud Vision API uses is available here. If you run the evaluation script in the repo you’ll get the EXACT same predictions as the Cloud Vision API.
6 Likes
Thanks for sharing Karthik, I had found the dataset a few days ago and was thinking about comparing the results!
From your experience, do you think that Clarifai and other ML API SaaS are using VGG16 or VGG19 ? or something similar ?
I was particularly impressed by their mention of needing only 15-30 samples per class to get good results.
Some kind of transfer learning I guess but I don’t know if or how it is superior to other approaches.
15 samples sounds very good to me for quick prototyping to get a quick idea of whether CNNs should be used or not.
I really don’t know. And I’m not sure if it makes much of a difference at all. For my own applications I use VGG-16 or Inception models. If I’m running inference on a phone I would be using Inception or Mobilenets.
What is the difference between VGG-16, VGG-19 and Inception?
Pros and cons?
Bumping this.