can't reproduce results by provided Glance PASCAL Checkpoints

[tianxia-diyi]

Hi,
I downloaded model checkpoints of Glance PASCAL from https://filebox.ece.vt.edu/%7Eprithv1/checkpoints/glance_resnet-152_pascal_1_2hl_250/
And I used pytorch Resnet152 pretrained model and tested on several different images. But it didn't give reasonable results. For different images, it always gave the same results. I have 2 questions:

1. For resnet152, which features are you using, fc7? The author only talk about vgg16 in the paper, but he shares model checkpoints trained on resnet152 features with us, which seems strange.
2. What kind of resnet152 model is used: torch, pytorch, or caffe?
   It is meaningless to provide the model checkpoints without specifying the above points. To clarify everything, could you please share the images features on which you trained your model with us? Thanks