Malt - Model assisted labeling toolkit

The fastest way to turn videos into training data

43 followers

The fastest way to turn videos into training data

43 followers

Malt empowers hackers to create labeled object detection datasets fast! You can quickly navigate through a video to find objects to label, and most impressively you can use a pre-trained TflowLite model to automatically place a bounding box around images!

Free

Launch tags:

Developer Tools•Artificial Intelligence•Tech

Launch Team

Matthew Sessions

Malt - Model assisted labeling toolkit

Maker

I built this tool after spending too much time finding and labeling training images for Tensorflow. A common workflow while building new object detection models is to convert a video to a bunch of images, find the images with the object I am trying to label, train a model, then play the model back through a video to visually see what angles are not performing well. Malt incorporates this entire workflow into a single tool. It is particularly helpful to have the pre-trained model auto-label the images and then quickly adjust the bounding boxes that do not tightly fit the object. This project can be installed with pip through PyPi. This is a quick demo of the product:

I hope others will find this tool useful and will be able to provide feedback. I am launching Malt as a beta application. Version 1.0 will have a cleaned-up UI and a few more features.