Every time I load my model it’s pulling resent model from PyTorch. Is this a normal behavior or am I doing something different for this happen? Shouldn’t the model I downloaded and fine-tuned be enough for making inference.
Downloading: “https://download.pytorch.org/models/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth” to /tmp/hub/checkpoints/fasterrcnn_resnet50_fpn_coco-258fb6c6.pth
This may be happening if pretrained_backbone=True and the model cannot find the official pretrained weights in the usual path. If you just want to use your own weights you can set pretrained_backbone to False and just load your weights manually like this
model = torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained_backbone=False)
model.load_state_dict(torch.load(PATH))
torchvision.models.detection.fasterrcnn_resnet50_fpn(pretrained_backbone=False) still downloads model from pytorch smaller size then previous. Is there a way I can load it from my directory and not download it from pytorch