3 ms·

Show HN: Babbly – point the camera and learn a language

2 points by barty777 6y ago

barty777 6y ago
My colleagues and I have decided to publish Babbly Android app to help people make most out of the isolation by learning a new language. Babbly is one of our side projects we have been developing for the past few months. Our team has a lot of expertise in Computer Vision and we have always wanted to test if in-the-cloud real-time inference is possible with AI. Babbly can process up to 5 pictures per second which demonstrates this approach is more than viable. The main advantages of the in-cloud model are seamless AI model updates, reduced battery consumption, and much lower restrictions on the model size which results in better accuracy. The model was trained on ImageNet and OpenImages public datasets which are not perfect for this kind of computer vision task, so the accuracy is not stellar. We plan to work on this issue in the future :) We realize this app doesn't offer much feature-wise, but we think any form of fun/knowledge can be helpful during this pandemic. And yeah, any kind of feedback is more than welcome :)
lecarore 6y ago
Feels like a solution looking for a problem. As an avid language learner, i really see no need for this, nor could figure out who could have a use for it. Reminds me of google photo auto tagging or google lens, but without any real usefulness. Here's a completely different problem for the same solution : let people auto-tag their local photo library using your cloud classifier. Some utility software would tag photos one by one in a folder, adding the tags directly as exif data. You get their raw data to improve your model (without keeping it), they get free tagging. I'm thinking about the photographers with hundreds of thousands of photos in their local archive that would like to be able to search for keywords like "wedding", "portrait", "beach" ... without giving away all their pictures to google.
- barty777 6y ago
  Thanks for the feedback! Automated photo tagging seems like a good idea and we will certainly think about it more.

2 points by barty777 6y ago

My colleagues and I have decided to publish Babbly Android app to help people make most out of the isolation by learning a new language. Babbly is one of our side projects we have been developing for the past few months. Our team has a lot of expertise in Computer Vision and we have always wanted to test if in-the-cloud real-time inference is possible with AI. Babbly can process up to 5 pictures per second which demonstrates this approach is more than viable. The main advantages of the in-cloud model are seamless AI model updates, reduced battery consumption, and much lower restrictions on the model size which results in better accuracy. The model was trained on ImageNet and OpenImages public datasets which are not perfect for this kind of computer vision task, so the accuracy is not stellar. We plan to work on this issue in the future :) We realize this app doesn't offer much feature-wise, but we think any form of fun/knowledge can be helpful during this pandemic. And yeah, any kind of feedback is more than welcome :)

lecarore 6y ago

Feels like a solution looking for a problem. As an avid language learner, i really see no need for this, nor could figure out who could have a use for it. Reminds me of google photo auto tagging or google lens, but without any real usefulness. Here's a completely different problem for the same solution : let people auto-tag their local photo library using your cloud classifier. Some utility software would tag photos one by one in a folder, adding the tags directly as exif data. You get their raw data to improve your model (without keeping it), they get free tagging. I'm thinking about the photographers with hundreds of thousands of photos in their local archive that would like to be able to search for keywords like "wedding", "portrait", "beach" ... without giving away all their pictures to google.

barty777 6y ago
Thanks for the feedback! Automated photo tagging seems like a good idea and we will certainly think about it more.