Fast annotation of images using superpixels. The image is over segmented into superpixels using the SEEDS algorithm.

The project is under development and expected to change frequently.

• CMake 2.8.11 or newer
• Qt5 (Qt4 should work with modified CMakeLists.txt)
• OpenCV 3.0.0 with contrib packages

Before starting the problem, we need to create a file that will contain the labels. The contents of the file could look like this:

car 0 128 255
bike 0 255 128
pedestrian 0 255 255

The first entry specifies the name, that will be shown in the GUI. The 3 numbers represent the RGB colors of each label. Note: A background class and a missing class are automatically added. Their respective colors are (0,0,0) and (128,128,128), so you might want to have different colors for your own classes.

The tool can be started via the command line:

./annotate labels.txt <indir> <outdir>

There is an option (-n) that controls the number of superpixels:

./annotate -n 600 labels.txt <indir> <outdir>

If no option is set, the default number of superpixels is 400.

For more information you can also view the help:

./annotate --help

You can use the following actions while labeling:

• <LeftMouse> labels the superpixel under the cursor
• <RightMouse> labels as background
• <Space> toggles contours of the superpixels on/off

• fix TravisCI build
• dynamically load labels from file
• improve file management in general