How to compress the YOLOV2 to reach the performance like SIDNet by TensorRT ?

As the SIDNet demo shown in GTC2018 , https://devblogs.nvidia.com/large-scale-object-detection-tensorrt/, DarkCaffe and TensorRT can help the modified YOLOV2(SIDNet)to run 2X times faster with FP32,and reach 6X times faster with INT8 @ batch=1.

Can we use darkflow to do INT8 quantization by the Tensorflow and use the TensorRT to compress the YOLOV2 to reach the performance like SIDNet ?

Any suggestions are appreciated.