Hi there,
Does anyone know how to replicate the same normalization with Imagenet stats that Fastai’s dataloader does ?
In fastai.vision.data you find imagenet_stats = ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
Which means the first 3 is the mean of the RGB images and the second 3 is the standard deviation.
You can do (x - mean) / stddev to normalize then.
Edit: You will also find the normalize function in the fastai.vision.data module btw.
3 Likes