In the modern world, videos are a very common and rich source of data. Street surveillance in your neighborhood, recordings of conferences and sports matches, movies with Ryan Gosling – all these may contain valuable data for analysis. Thus, it is important to have efficient tools to label videos.
Since Hasty is a computer vision platform, we wanted to let the users work not only with image-based projects but also to address video-based cases.
To do so, we introduced a Video Annotation environment where you can:
When creating a new project, toggle on the “Video-based project” option:
Upload videos you want to work with. You can add several files.
One of the typical tasks video annotators work with is Activity Recognition. The idea is to identify a sequence of frames that contains a certain activity.
For example, if you are doing football (or soccer) analytics, you might want to quickly detect activities such as passing, dribbling, and shooting. With the Activity Recognition feature, you can also train a model that can detect when team A has a scoring opportunity in the football match recording.
The workflow for Activity Recognition is the following:
Create the activities you want to detect in the video:
Label the activities by creating segments throughout the clip:
If you confused the activities, you can always edit the segment and its length.
Activate the AI Assistant: when you have finished labeling your clips, set video status as “To review” or “Done”. This will train the AI assistant. Once unlocked, it will start suggesting the potential activities for labeling;
Annotate the rest of the project with the help of the AI assistant.
If you work with several files, you can always switch between them via File manager.
When labeling is over, you have 2 scenarios that depend on the goal you pursue:
1. Train the model that extracts activities: export the activities to JSON so you can experiment with the model outside Hasty.
2. Extract frames for image-based annotation project: export the activities and segments to the Hasty image project to use all the Annotation Environment capabilities to label the interesting footage in detail.
The Video Annotation environment in Hasty allows you to work with video-based projects.
Within it, you can:
Create a model that will detect and label certain activities in the videos & export them to JSON;
Export segments with detected activities into image projects and work with frames.