what do you want to achieve here? ROI pooling/ align/ wrap are just preprocessing techniques, if you want an algorithm for object classification, detection and instance segmentation have a look at these repositories:

SOTA: https://github.com/joe-siyuan-qiao/DetectoRS

Mask RCNN: https://github.com/matterport/Mask_RCNN

YOLO4 (for object detection only): https://github.com/Tianxiaomo/pytorch-YOLOv4