Semi-Automatic Video Annotation @ Scale for Machine Learning

It is no secret that the amount of video data being generated by diagnostic and monitoring systems alone is very large. For machine learning teams, it is difficult to pre-process such large quanitities of training data.

To enable fast annotation of large volumes of video content, TrainingData.io has launched a semi-automatic video annotation tool that augments the work of human-labellers.

Let's assume you have a video of multiple objects moving accross the screen. As an example take this video of nascar cars racing. Select the objects that you wish to annotate in the first frame. Select button "Track Object". Wait for the results to be generated. Inspect the results and fix them if necessary.

The objects do not have to be of same type or any specific type. The object detector is not a neural network or a DNN.