When running the video data of our own, it was found that the appearance contours of the generated objects were always uneven. Is there any way to improve it?

[bangbangDong]

The original picture is

Or when building a complex model, some of the objects become blurry：