This post will give a high-level tour of how to create bounding boxes for object detection requirement using the CVAT(Computer Vision Annotation tool). The prime inputs to any object detection model are the image and the location of the objects in the picture(via bounding boxes). We have a wide range of tools to create the bb boxes, but for this discussion, our choice is inclined towards CVAT.
After creating an account in CVAT.org. The next step is to create a task to start the process of placing the boxes.
By clicking on the ‘Create new task’ button, the user will be directed towards the place where the details of the task need to be provided. One of the critical pieces of information is the labels for the objects in the image. For instance, let's take the below picture, we’ve four unique objects(orange, coolers, plate and a glass). Every object will be labelled with a distinct name.
Providing the details of the objects during task creation. We can see an ‘Add label’ button via which the classes can be included. This is followed by the target image.
After submitting the details, we could notice a new task added to our list. We can click on the open and choose the job to start our annotation process.
In order for a simple bounding box, we can click on the rectangle shape and choose the respective label for the object, then draw a rectangle over the corresponding objects.
By repeating the same steps, we can create bounding boxes for all the objects with the respective labels.
Once the bounding boxes are created, we can save the changes. For the object detection models, the bb box details can be downloaded by selecting the appropriate option based on the architecture employed.
Please refer to the user guide below to explore many other techniques and the usages.