I want to extract features using a fine tuned vgg16 model based extractor. I want to know, should i extract features in model.eval() mode that is I should put my extractor model to eval mode first and then apply it to input image for feature extraction? or should I use my extractor with in with torch.no_grad():