Question about FCNResnet50 pretrained weights on torchhub

I wanted to use a pretrained FCNResnet50 model to run segmentation task on Pascal VOC dataset. However, when I import the fcn_resnet50 — Torchvision main documentation ( weights and run inference on Pascal VOC validation dataset, I get nowhere near to the performance that is described here.
So, my question is whether the above mentioned FCNResnet50 actually trained on Pascal VOC dataset or am I missing something ?