Lobe.ai — Drag and Drop AI tool acquired by Microsoft
Overview:
Lobe is an amazing and powerful tool for AutoML. It has everything you need to bring your machine learning ideas to life. Just show it examples of what you want it to learn, and it automatically trains a custom machine learning model that can be shipped in your app.
Lobe is a Windows or Mac desktop software program that allows everyone to create machine-learning models for image classification. It lets you build machine learning models with the help of a simple drag-and-drop interface.
So, everyone has access to machine learning skills and can create models without needing any technical knowledge.
The steps for training a model with a lobe are very simple:
· Create a dataset using a webcam or existing images
· Assign the labels
· Train the model
· Evaluate outcomes
· Export the model
Currently, Lobe supports only image classification projects, but in the future, lobe will come with Object Detection and Data Classification as well.
For this article, I have used Yoga Poses Dataset. There are many yoga poses but the very well-known ones like downward dog pose, goddess pose, tree pose, plank pose, and the warrior pose. Using lobe we will try to classify images among these six classes.
Playing with Lobe:
The first step is collecting the data to train the model. In lobe, 5 samples per class are needed to start training the model.
There are three methods for Data collection:
1. Import images from your local computer
2. Capture images using a webcam
3. Import a structured folder of Images
In this example, we will import a structures folder of images so it will automatically assign the label to each image according to the folder name. I used this method to save our time when doing data labeling: Instead of just throwing a bunch of photos into Lobe, dedicate a few minutes to create folders and name them using the Labels we want to use in our model. Then put the images in their corresponding folder.
The best thing about Lobe I like is that it pretty much does Building the Machine Learning Model and Training the Model for us. After importing and labeling the data lobe starts building a model and training.
As you can see on the left, I ended up with 1036 images. This is because I wanted the model to be even more precise than what I had in the first iterations.
In those first iterations, the model worked surprisingly great. But it did have issues sometimes recognizing between downdog, goddess, and warrior and the level of confidence of its predictions was in some cases quite low. Bringing more images to each label made the prediction much more precise.
In lobe, we can correct the wrongly classified labels and again train the model. So it will become more accurate.
Play with predictions:
Lobe already comes with a “Use” section in which you can test your trained model. In the Use section, you can drag-drop new images or take photos with your webcam. Lobe will run the trained model with this new image and you can see how good your model does with the new images.
Another great feature of the lobe is that you can try to trick your model and see patterns where it is weak. You can also help improve your model by giving feedback on its predictions by clicking the checkmark button to add the image to your dataset. Lobe automatically trains the model with these new images. If the model predicts the wrong label to an unknown image, you can change the label at the testing time and again give that image to train the model.
Exporting the Model:
When you’re satisfied with your model prediction, you can export the model to a variety of industry-standard formats and ship it on any platform you choose. Also, you can easily optimize the trained model using lobe.
Here is the list of some platforms where you can easily ship your model.
· Lobe Connect
· TensorFlow.js
· Web App (React)
· TensorFlow
· ONNX
· TensorFlow Lite
· REST Server
· iOS App
· Android App
Features of Lobe.ai:
Image Classification:
So far Lobe only has one main feature and that’s training and image classification network. And it does that pretty well. In all the tests I have done I have gotten decent results with only very little training data.
Speed:
The speed is insane. The models are being trained in something that seems like a minute. That’s a really cool feature. You can also decide to train it for longer to get better accuracy.
Export:
You can export the model by a local API, CoreML, TensorFlow, TensorFlow Lite, and many more.
Comparison with other AutoML solutions:
Fewer insights
In short, you don’t get any model analysis as you would with Kortical for example.
Fewer options
As mentioned Lobe only offers image classification. Compared to Google Automl, which does that and objectregocnition, text, tabular, video, and so on, it is still limited in use cases.
It’s Easy
This is the whole core selling point for Lobe and it does it perfectly. Lobe is so easy to use that it could easily be used for teaching 3rd graders.
It’s Fast
The model building is so fast that you can barely get a glass of water while training.
Quality
When I compared a model I build in Google AutoML to one I build in Lobe, Google seemed to be a bit better but not by far. That being said the Google model took me 3 hours to train vs. minutes with Lobe.
Conclusion:
In conclusion, Lobe.ai is a great step forward for accessible AI and already in its beta it’s very impressive and surely will be the first in a new niche of AI.
It doesn’t get easier than this and with the export functionality, it’s actually a very good candidate for many commercial products.
