Wrong inference result with python API for tensorflow model

Hi, I am using a frozen tensorflow version of YOLOv2. The problem is when I used tensorrt with it, the inferred result is way off, with almost all of the result are -inf.

I am using tensorrt on GTX 1050, tensorrt 3.0.4 with cuda 8.0. Any help would be appreciated.

The problem turns out to be some error with setting up some code.