[#153] add epsilon number to avoid different result at different scale

Cartucho / mAP

mean Average Precision - This code evaluates the performance of your neural net for object recognition.

Apache License 2.0

2.93k stars 910 forks source link

[#153] add epsilon number to avoid different result at different scale #154

Open ptran1203 opened 3 years ago

ptran1203 commented 3 years ago

I check your code and see that you + 1 for intersection width and height. I think you want to avoid zero division error, but it leads to a significant shift if I measure the results at different scales (image resolution).

I think we should replace 1 with a very small number such as 1e-6

https://github.com/Cartucho/mAP/blob/master/main.py#L557..L562

See Issue #153