A little rant, a little show me how I can help?
Docs show here a denormalize function and below that, state
The imagenet norm and denorm functions are stored as constants inside the library named
imagenet_norm
andimagenet_denorm
However, release notes say that’s actually incorrect post 1.0.14. They have been removed.
Only way I could find to denormalize is this:
learn.model.eval()
x, _ = next(iter(bb_db.valid_dl))
preds = learn.model(x)
_, denorm = normalize_funcs(*imagenet_stats)
im = x[0]
im = denorm(im)
im = to_np(im).transpose(1,2,0)
fig, ax = plt.subplots(1)
ax.imshow(im)
That the expected way? Create function from stats used for databunch?
Worth updating the docs to mention that the supplied imagenet_norm and denorm don’t exist?