인프런 커뮤니티 질문&답변

안녕하세요 gaussianblur 이미지 augmentation을 적용할 때, 이미지에 cuda에 붙였을 때와 안붙였을 때 수행시간 차이가 발생하는데, 이유가 있을까요?

아래 코드에서는 0.001s, 0.001s씩 걸리는데

import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision
from torchvision import transforms
import time
import albumentations as A
import numpy as np

sample_img = torch.randn((1, 3, 224, 224), dtype=torch.float32)
sample_img2 = torch.randn((3, 224, 224), dtype=torch.float32).cuda()

gaussian = transforms.GaussianBlur(5)
since = time.time()
gaussian_img = gaussian(sample_img)
print(f'time: {time.time()-since:.3f}s')

since = time.time()
a_gaussian = A.GaussianBlur()
gaussian_img2 = a_gaussian(image=sample_img2.permute(1, 2, 0).detach().cpu().numpy())['image']
print(f'time: {time.time()-since:.3f}s')

위 코드에서 sample_img에 cuda를 붙이면 오히려 시간이 0.4초로 늘어나게 되네요