A simple video browser with Python and Tkinter. Go through video frames and mark the interesting sequence begin and end points. Uses opencv for the video reading
- Load annotations from file
- Save annotations to file
- subclass the annotator and override the part for displaying detected objects to annotate any shape or format object detections (pose, bboxes)
- reannotate already existing detections or annotations with additional classes
The only dependency of PyViAnnotate is opencv. This is developed with opencv 4, but will probably work with opencv 2.4 and 3.* as well since only core features are used. Opencv python bindings are found on the PyPI on most platforms, no need to start installing it separately.
The directory is installable as python package. On some platforms a separate installation sudo apt install python3-tk might be necessary.
cd pyviannotate
python setup.py installpython3 annotate_video.py --video /path/to/video.mp4A small demo picture. GUIs made with Tkinter have a professional look from the 90s.
Annotation files are stored as json. The objects are stored for each frame. The coordinates of detected objects are stored as a list of coordinates. This allows storing more complicated shapes that just boxes. For bounding boxes, two coordinates are stored, the upper left and the lower right corner. A unique object id is also stored for each detection. It can be used for tracking the object through frames.
{
"frame_count" : 1,
"frames": [
{
"frame_index": 0,
"objects" : [
{
"class_name" : "class1",
"class_id" : 1,
"object_id" : 0,
"object_coords" : [
{
"x": 123,
"y": 123
}
]
}
]
}
]
}