yolov7-Pose compared with edgeai-yolov5

[StochasticGame]

I put your recently submitted YOLOv7 pose detection model into my own modified edgeai-yolov5 project and used the inference prediction in the detect.py script, and the inference results were very good, but I had the problem of the key point going out of the target detection frame when using the model in edgeai-yolov5 for inference, I would like to know I would like to know how your yolov7-pose was trained, or how you avoided the keypoint out-of-bounds problem（yolov7-pose：Figure 1；edgeai-yolov5-pose：Figure 2） Figure 1 Figure 2

edgeai-yolov5/issues：https://github.com/TexasInstruments/edgeai-yolov5/issues/16

[WongKinYiu]

I do not do anything to avoid it. Just put yolor pose to edgeai-yolov5 pose code, and use same training script to train yolov7 pose.