Train problem

[Lin-Zhipeng]

When I trained my own dataset,I find the test result of all my object have negative confidence score and cannot identify the extreme point on the left very well. Have you ever had a similar problem on your side, or did you know what might have caused it.
Demo result

[xingyizhou]

The link is invalid. There should not be confidence scores. Can you specify where you inject your visualization code?

[Lin-Zhipeng]

I just run the demo.py to visualization results. I set the threshold is very small.
You can copy the link to browser to see the picture.