SSAT Anotation Tool
SSAT is presentely an in-progress project of a free and interactive anotation tool for the computer vision community, it eases the way reseachers can annotate data from massive video datasets.
Note that for the moment the annotation tool has only versions for MAC and Windows, to install the tool in this two operating systems
visit this link and dowload the application installer
If the user prefers to compile the source code there are two major dependencies, the user will have to install OpenCv 3.0 or 3.1 with contrib libraries, and then execute the cmake file in an environment that has qt libraries.
There are a few requirements thata make the SSAT work properly.
No operation can be executed if no videos has been previously loaded, to load a video press “File” in the menu bar and then “Open” (or use the sortcut CTR + O)an open dialog will pop up and you can select any video that has a format .mp4 or .avi Note that in the image all features are locked because no video has been loaded After the video is loaded, all functionalities start to work properly
- Before begining the annotation the user must establish the atributes in wich he desire to base his annotations , to do this select “Annotation” in the menu bar followed by attributes, another dialog will pop up and the user can create an hierarchy of categorys(filters) and labels. The attributes box, showed in the image above, will be initialy empty, the attributes are defined following an hierarchy of categorys and labels, this hierarchy was created to ease to use of the tool in the midst of a high number of different labels. Hence a category can be understood as an collection of labels, that have something in common. It is not possible to create a label that doesn´t belong to a category, so the “insert label” button remains deativated until an category is selected, the user can also edit an annotation already made by double clicking it on the table and making the desired changes.
- There are two tipes of annotation: frame based annotation and bounding-box based, the former comprises the whole video frame
whereas the latter specifies a region(box) from the video picture, to annotate right-click on the video and then choose between frame based and bounding box based..
- Frame boxes hold the following info: category, label, initial frame, final frame and general information.
- Bounding boxes encompass information like: category, label, initial frame, final frame, x-axis position, y-axis position,width, height, and general information.
Presently, the bounding box based annotations are still being developed hence, they do not work properly, the user can create boxes on the screen and they are saved, but the bounding boxes can’t be removed or changed in the core of the SSAT
- Whenever the user decides to import/export a file he must click on file and then selects import/export JSON (or use the shortcut CRT + E to export and CRT + I to import) and then specify either desired the file to import or the folder to export. Note that every 10 seconds a security file will be exported for the root directory and if the user open a video he already has opened before, he will be asked to import this security copy.