Hi, I would like to know where the parameters in this transform come from? I don't get the same values if I take the mean and std over the images 'train+unlabeled' split.
https://github.com/untitled-ai/self_supervised/blob/6d14ca0402ecc13feda9b3a9fdc056fd1ac24473/utils.py#L127-L129
I tried the method used to get these values for ImageNet in torchvision described here on various subsets of train+unlabeled and the std gets closer but the mean is still off.
I am resizing first to 128 then center cropping to 96 as you appear to do for validation images.
I will note that I am using TensorFlow and perhaps the resizing method might be having an effect. However I also tried using PyTorch and still could not reproduce these values.
Thanks
Hi, I would like to know where the parameters in this transform come from? I don't get the same values if I take the mean and std over the images 'train+unlabeled' split.
https://github.com/untitled-ai/self_supervised/blob/6d14ca0402ecc13feda9b3a9fdc056fd1ac24473/utils.py#L127-L129
I tried the method used to get these values for ImageNet in
torchvisiondescribed here on various subsets of train+unlabeled and the std gets closer but the mean is still off.I am resizing first to 128 then center cropping to 96 as you appear to do for validation images.
I will note that I am using TensorFlow and perhaps the resizing method might be having an effect. However I also tried using PyTorch and still could not reproduce these values.
Thanks