Thank you so much for your suggestion. Through a lot of troubleshooting, this issue has been solved. The solution sounds funny. Because I use to label frames using labeling in small dotsize, e.g. 1, thus in the labeled videos, the labelings are also very small, which is almost unseeable. Once change to larger dotzise, everything works normal.
Also, because I combined multiple images (multiple camers) into one, so the images are very large. Right now, DLC works very well to process these large images. Do you have any other good ideas to analyze these large images or is there any potential problem of using large image?
Thank you so much!