I am trying to use torch vision API for faster RCNN and use an existing pretrained model trained on large number of general class labels such as ImageNet. But the official docum