Can I use ROI Pooing to extract region features from a feature map of the entire image in any task?

Question

Is ROI pooling (e.g. in Pytorch) suitable for any tasks instead of object detection? For example, a pre-trained VGG16 on ImageNet. I use this image example as the input with