Hello Superxiong,
The problem is that there are almost no tie points detected on the banana's surface, which seems to be a reason of the image quality. Try to use better camera or at least provide better lighting (do not use flash, but have some diffuse sources of light instead), also use the image frame space more effectively.
The sample dataset that we have provided for the tie point masking is also far from ideal and is not considering many of the recommendations of image acquisition, so you do not have to follow the same visual example.