Label_BoundingBox

A tool for labeling bounding boxes of training images. It is based on BBox-Label-Tool and supports multi-class labeling. Here is a demo:

Requirement

• Python (2 and 3 is all ok, 2 need a little change)
• OpenCV (It isn't necessary, if you only use labeling function. OpenCV is only for getFrameFromVideo.py)
• tkinter
• PIL
• JSON

Usage

for basic bbox-label-tool:

• Add paths to images (imgDir) in "label_boxes.py"
• Modify how many times smaller the images will be scaled (rescaleFactor) in "label_boxes.py" if necessary.
• Start label-tool with "python label_boxes.py"
• click "load" and you can label images now!
• To create a new bounding box, left-click to select the first vertex. Moving the mouse to draw a rectangle, and left-click again to select the second vertex.
• To cancel the bounding box while drawing, just press "<ESC>"
• To delete a existing bounding box, select it from the listbox, and click 'Delete'.
• To delete all existing bounding boxes in the image, simply click 'ClearAll'.
• After finishing one image, click 'Next' to advance. Likewise, click 'Prev' to reverse. You can press "<Left>" and "<Right>" Or input the index and click 'Go' to navigate to an arbitrary image instead. The labeling result will be saved if and only if the 'Next' button is clicked.
• Individual JSON labels are stored in the images folder.
• As a precaution, a new JSON file will not be created if the file does not exist.

for multi-class labeling:

• Write down your classes in "class.txt"
• Start label-tool and select class at the top-right corner. remember to click "ConfirmClass"
• After clicking "ConfirmClass", you can label different classes with different colors.

for get Frame from Video:

It isn't user-friendly. If you want to use it, I recommend you to read source code carefully!