Model Assisted Labeling Toolkit

Malt allows users to quickly create labeled images from a video with the help of a predefined model. Creating good training data for object recognition is an iterative process, and malt is designed to speed up each iteration. The typical malt workflow will go as follows:

• Create a sparsely trained Tensorflow Lite Model
• Load and scan through a video to find object positions that are unfamiliar to the model and label those frames
• Adjust the auto-generated bounding boxes to perfectly fit the target object
• Save the labeled frames as images
• Train a new model
• Load the new model and a new video
• Repeat the process

Install & Launch Malt

pip3 install maltk malt or python3 -m malt