Microsoft has released a program that allows users to perform deep learning on images without the need for programming knowledge.
The app consists of a simple interface that, in three steps, lets the user make the software recognize and add custom labels to images. The steps are explained in the video below.
Label, train, play
The software opens with the first step, called Label. In this step, the user adds a couple of labels in which he wants to divide the images. For each label, the user then adds at least five pictures to train the algorithm with.
Next, under the Train tab, the user is shown what the algorithm has made of the images and which edge cases still confuse it.
When the algorithm has finished training, the results are shown in Play. There, the user can add more pictures to improve the algorithm if needed. Finally, the results can be exported from Lobe to be used in CoreML, TensorFlow, TensorFlow Lite or a local API.
Lobe was developed by a startup of the same name, which was acquired by Microsoft in 2018. The software has already been used by Nature Conservancy to analyze pictures made by tourists in the Caribbean. The algorithm was tasked to show where and when visitors interacted with the coral reefs.
The preview version of the software can be downloaded for free from Lobe’s website.