This tool was created to interact (visualize, create, modify, remove) with Bonn Activity Maps[1] datasets. Additionally, the tool supports interactions with Posetrack[2] datasets. The following sections contain the explanations of the main functionalities of the tool.
The tool has a user management system and is therefore protected by a login. This system allows two different types of user:
The tool allows to have up to 4 camera views simultaneously to interact with. This views can be switched at all times with the ones stored in the views storage panel situated on the right of the tool. The storage panel hides when not being used.
Since all the views are synchronized, moving through the timeline will update all the views to match the actual frame.
The tool allows different types of objects; "Persons", "Poses" and "Boxes". Each object has an individual and unique UID for fast identification. Multiple objects can be visualized at the same time. (TODO: maybe show the process focusing in a single person and show all assigned object types and actions to that UID)
In the Keypoint editor you can visualize which frames (top row) have annotations for each object (first column). There are also special operations that can be perfomed at an object level and for a range of frames(red buttons):
Each object can be individually edited for each individual frame:
In order to speed up the annotation process several operations can be performed:
Saving the annotation interpolates automatically (this option can be disabled in the Options Panel that will be shown later).
The objects of type "Box" are considered static and their behaviour is different. They represent entities whose position and orientation barely change during the whole recording, for example, chairs, tables, fridges, etc.
Each "Box" object has a label field to mark them as specific real-world objects (chair, table, fridge, etc.). When creating/editing a static object, changes will be replicated forward in the timeline, so all following frames will be also updated.
Since the datasets expect the annotations to be in 3D, the tool provides some aid to ease the process. When a keypoint is placed in one of the views, epipolar lines will appear on the other views. This lines can be used as a reference of where should the keypoint lie in that view. Once a point is placed in at least 2 different views, we can save it. The tool will then triangulate those points and generate the final 3D point.
Users can also annotate the actions being performed by each "Person". This actions can be visualized directly on the views.
In the Actions Editor Panel, new actions can be created, specifying action type, start frame and end frame, and deleted.
The tool lets the user change some options so that everything fits his/her specific preferences.
Almost every action that the user can perform has a keyboard shortcut to speed up the annotation process. This shortcuts are available for different keyboard layouts to fit everyone.