Let say I have a task to detect the bounding box of one object only. And the only thing I care about is the IoU between prediction and ground truth, no need for real-time.
- Should I should one-stage (YOLO, SSD) or two-stage (Faster RCNN) detection model?
- If two-stage is better? If I use Faster RCNN, should I train the whole model or only train the Region Proposal Networks?
P/s: I have not read some state-of-the-art model yet, I heard that they implemented some really interesting things in EfficientDet or DETR, if you think those models better than two of my suggestions, please let me know.