Low mAP in inference by using the provided model

[hangxu124]

Hi, Thanks for your effort in making this yolo_v2 TensorFlow model.

Have you ever tested using ./test/03.jpg and ./test/02.jpg with VOC model?

I did the inference on these 2 images and found that the performance (bbox) is really terrible... So I just wonder that if I did sth wrong or the models do have bad performance...(maybe not well-trained?) I also wonder how about the mAP on COCO2017 or VOC2007 dataset with the provided model?

Thanks!