Open yangninghua opened 3 hours ago
Hello @yangninghua ,
Thanks for reaching out to us.
While fn.erase
itself doesn't allow for any complex patterns, you can fill the union of shapes covered by erase
with contents of an arbitrary tensor of matching size.
mask = fn.erase(fn.zeros_like(image), fill_value=255, < your mask shape parameters >)
noise = fn.random.normal(image, mean=..., stddev=...)
out = (image & (255 - mask)) | (noise & mask)
Describe the question.
Does Dali have plans to implement the per_pixel mode?
https://docs.nvidia.com/deeplearning/dali/user-guide/docs/operations/nvidia.dali.fn.erase.html#nvidia.dali.fn.erase
Check for duplicates