I am wanting to use create bounding boxes in order to crop images before uploading them to Datarobot. What python/other software would you recommend to achieve this?
Just for context looking at the DR activation maps (shown below) of one of the images most of the pixels aren't used in the classification of the image so to reduce the projects size Ideally they can be cropped out before modelling.
Solved! Go to Solution.
Just making sure you got everything you needed on this. Since the original question, many of us have experimented with combining Yolo v5 with DataRobot for bounding boxes. Please reach out to your account team if you'd like a demo.
Ah great thanks @anton.kasyanov, I hadn't heard of yolo or CVAT checking them out right now. The reason I'm concerned by DataRobot rescaling is I assume in order to do this without cropping causes the image to become pixilated as the information is compressed to a smaller size thereby reducing accuracy.
Are you sure you need to crop them? The fact that DataRobot does not use full image does not mean that you need to do anything extra. Unless you are sure that your model will really benefit form it.
If you still decide to crop objects you can use any open source object detection models like yolo or tools like CVAT.