Describe the question. I haven't found a suitable API to perform i

Hi <a class="user-mention notranslate" data-hovercard-type="user" data-hovercard-url="

How to use Dali tool for image sharpening operation about dali HOT 4 CLOSED

wangdada-love commented on May 30, 2024

How to use Dali tool for image sharpening operation

from dali.

Comments (4)

JanuszL commented on May 30, 2024 1

Hi @wangdada-love,

Thank you for reaching out. If you want to add images with weights please just:

imge3 = img1 * weight1 + img2 * weight2

from dali.

wangdada-love commented on May 30, 2024 1

Thanks for your reply.
I have tried this method before, But I received an error message as follows:

RuntimeError: Critical error in pipeline:
Error when executing GPU operator conditional__Merge, instance name: "__Merge_29", encountered:
[/opt/dali/dali/pipeline/operator/builtin/conditional/merge.cc:45] Assert on "base_input.type() == input.type()" failed: Divergent data found in different branches of conditional operation. All paths in conditional operation are merged into one batch which must have consistent type, number of dimensions, layout and other metadata. Found distinct types: float and uint8.

I just tried again, but got the same error message. Based on the error message, I speculate that the problem is caused by an incorrect data type. So, I made the following modifications to the code:

def rand_sharp(images, scale=1.0, window_size=3, device="gpu"):
    sharp = fn.laplacian(images, scale=scale, window_size=window_size, device=device)
    sharp = fn.cast(sharp, dtype=dali.types.DALIDataType.FLOAT)
    images = fn.cast(images, dtype=dali.types.DALIDataType.FLOAT)
    out = 0.7 * images + 0.3 * sharp
    out = fn.cast(out, dtype=dali.types.DALIDataType.UINT8)
    return out

it works.

from dali.

JanuszL commented on May 30, 2024 1

Hi @wangdada-love,

I'm happy you made that work. The issue is that out = 0.7 * images + 0.3 * sharp returns float type while the other branch of your if condition returns just int8, so in your context just

def rand_sharp(images, scale=1.0, window_size=3, device="gpu"):
    sharp = fn.laplacian(images, scale=scale, window_size=window_size, device=device)
    out = 0.7 * images + 0.3 * sharp
    out = fn.cast(out, dtype=dali.types.DALIDataType.UINT8)
    return out

should do as well.

from dali.

wangdada-love commented on May 30, 2024

outstanding! thank you very much.

from dali.

Recommend Projects