This repository implements a bounding box annotation tool
pytorch (preferably with cuda) ultralytics pynput opencv numpy
Make sure to add your images (your zip file) to the 'captured images' folder, and rename it to 'data' Also, install all dependencies Now you can run the main.py file
First step is to check / draw the bounding boxes. A model is supplied that suggests bounding boxes.
The two steps below can be repeated until you press 'enter' when all bounding boxes are visible (which means you are done and can go to the classification step!)
- Look at the results: press 'e' to edit
- When done or when you want to check, press 'esc' To draw a bounding box, press your middle mouse button (wheel). To zoom in, roll the middel mouse button (wheel)
The second step is to classify each bounding box Use the left and right arrows to switch between selected labels, the selected label is visualized in the terminal Press 'enter' to label selected bounding box with the selected label Press 'c' to not assign a label (= faulty bounding box) Press 'z' to undo your previous label, if you made a mistake Press 'esc' to finish the annotation early when you are sure the other bounding boxes are faulty