Pre-trained Model
PaperKites opened this issue · comments
Is the pre-trained model trained on the Places2 or ImageNet dataset?
Because in the net.py file, a pre-trained model is called (vgg16) and it's trained on ImageNet.
And there is a python file called Places2.py, so I'm not sure on which dataset is the 100000.pth trained on.
Thank you for your time.
Since the model is from torchvision, it is pre-trained one ImageNet. Please make sure to visit docs of torchvision.models.
So for the 10000.pth model, did you train it on the Places2 dataset? Because if so, it might be the cause of the artifacts (patches) that happen in the image corners. Because ImageNet has a different image size (resolution) than places2 dataset
PS: I'm not an expert, but I thought that might help ^^
I guess all the images are resized to a fixed resolution, regardless of the dataset used for training.