I am trying to use the given vgg16 network to extract features (not fine-tuning) for my own task dataset,such as UCF101, rather than Imagenet. Since vgg16 is trained on ImageNet, for image normalization, I see a lot of people just use the mean and std statistics calculated for ImageNet (mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]) for their own dataset.
Now I am confused. If I want to extract features with VGG16 (pretrained on the ImageNet) on my dataset, should I subtract the ImageNet mean or should I calculate my dataset’s mean and std firstly? Is it necassary?Are there big difference between the ImageNet and other RGB datasets generally?
Could anyone help figure it out? Thanks!