What is the difference between a bounding box and ROI (Region of Interest)

Question

I was reading about the Fast RCNN for object detection. From what I understand, it uses pre-computed ROI's (using selective search) and uses these to predict the bounding box offsets and uses smooth L1 loss to refine these and get closer to the ground truth boxes.

The paper states the following about the ROI's

While training, R/N ROI's for each image (N=2,R=128) are taken where N are the images per mini batch. Among the ROI's chosen, around 25% of them are taken that have IOU more than 0.5 (foreground) and the remaining (called background) that have IOU between 0.1 and 0.5.

My question here is, are these ROI's similar to bounding boxes and if not, in what way are they different?

Any help would be appreciated. Thanks!

Answer 1

A region of interest is a patch of the image which is sent to a classifier; it may not match the ground truth object bounding box. The bounding box prediction is computed from the features in the region of interest (e.g., via a linear regression or neural network regression) and should more closely match the ground truth bounding box.