Lobe

@merlin_laffitte hey Merlin! Thanks for your questions! We really hope so! 1. We are planning on expanding to other problem types soon. The first one will most likely be object detection—so we will still be in computer vision and image recognition, however Lobe is built on top of a solid foundation that will allow us to expand to text, tables, numbers, and more in the future. 2. Learning happens in two ways in Lobe. One that is fixed-epoch based—which allows you to play with your model and iterate on it pretty fast. The other one is also epoch based, but Lobe trains your model until it fully converges. This is what happens at the end of the video (8:06), when Jake let's his model Optimize. Again, here to help if you have any other questions, and thanks for the comment :)

@merlin_laffitte @ramongilabert Very cool! What about audio?

@merlin_laffitte @ramongilabert @michael_hughes2 we are thinking about audio too! for now you can convert audio into a spectrograms and use image classification to make predictions. Do you have any audio projects in mind?

@mbeissinger hey Markus, thanks for this great product ! 2 quick questions: 1/ can we use the trained model in another application ? How ? 2/ what's the pricing? Thanks !

@b_maurin yes you can use the trained model in a variety of places -- we export to the standard formats of integration like Tensorflow, TFLite, and CoreML (https://docs.lobe.ai/docs/export...). We also have some starter projects in code for running on mobile, the web, or spreadsheets of image URLs in our GitHub: https://github.com/lobe/ Lobe is free and runs locally on your computer :)

Hi @moproduct @mbeissinge! Wooooowwwww! I am amazed, so easy to train and ship ML Models, I just saw the drinking water example and I was so shocked! This is the future! Kudos to the team for creating such a great product! Wow! Congratulations!

@moproduct @porush_puri thank you!

@johnbhiggins let us know when you use version 2! This one is available to download right away at lobe.ai

@johnbhiggins @ramongilabert I'm sure the answer is "can't say" but will this be replacing azure machine learning studio (another product I love)? I recently saw in the serp title that that "(Classic)" was appended to the end of the title.

@johnbhiggins @ramongilabert you are right, the answer is I can't say 😋 but I wouldn't worry too much about that.

@johnbhiggins @ramongilabert a revised version of that product recently became Generally Available as Azure ML Designer and it is still alive and well! Pretty different from Lobe. Try them both!

@ay8s thanks Andrew! are you having issues labeling large numbers of images on Mac with Lobe? A few tips: 1. You can drag in a folder of organized images and Lobe will automatically label them for you (folder name = "Cats" with 500 images, they will all be labeled "Cats") 2. I see you're an iOS dev :) We have an iOS bootstrap app on GitHub to make it super easy to export a trained model to an iOS app! https://github.com/lobe/iOS-boot... Let us know what you think!

If there are only two labels like Drinking, Not Drinking I wonder if the UI for labelling could be optimized to simply be a Checkmark/Cross like the Play section. That way it would be even fewer clicks to label more images.

@cohenjacobd Thanks, will take a look. I imported a unorganized folder of about 8000 images so I'm not surprised it had a few issues. A few images didn't import successfully so the failed dialog showed up but it was pretty sluggish showing/dismissing. Looks awesome though! :)

@ay8s thanks for the comment Andrew! And yes, we are really excited about that, too. We are working on performance to support larger datasets as we speak as well! Stay tuned on our Twitter channels or on lobe.ai 😊

@cohenjacobd @ay8s ahh got it okay good to know! Today we support PNG, JPG, BMP, and WEbP formats. What image formats are you importing? We will be adding support for more formats over time. You can learn more from our docs here and get tips for training a model and exporting too! https://docs.lobe.ai/docs/welcom...

@dannypostmaa wow, I love this idea. I started doing something similar with typography, I have a model with +1500 images that tells you the difference between different type categories. Can’t wait to check out your project.

@dannypostmaa @ramongilabert Oh man I should add that too!

@dannypostmaa @ramongilabert I don't think I can post links in here, but if you search on Twitter for lobe_ai, there is a person that did a model classifying Hebrew letters as well. What I did to create the serif/sans serif/traditional/grotesque/etc. dataset was to add artboards of a good size in Sketch, take 20 or 30 different fonts, and put all the letters I could. After you create the first row, you can duplicate it and change the type pretty easily. I ended up with around 2000 images in pretty much no time.

@aidy_love thank you! We are working on object detection as we speak, and we hope to bring it really soon. I can't give you an estimated time yet, but know that we are really committed to it. Data classification comes after, though the foundation for it has been built 😊

@ethanyfan hey Ethan! We are purely using image classification, and have built a similar example counting pushups as well with Lobe. You're thinking in the right direction though - in the future we would love to have the model architecture chosen based on a more detailed understanding of what your data represents.

@ethanyfan @mbeissinger it’ll be great to have that option, but for many scenarios simple image classification gets the job done faster and more easily. Helping guide people in decisions like this will be an interesting challenge!

@ethanyfan @mbeissinger good to know.