import numpy as np
from PIL import Image
import pytest
import torch
import torchvision
from torchvision import transforms as trans
import cv2
import torchvision_npu
def custom_crop(img, pos, size):
ow, oh = img.shape[:2][::-1]
x1, y1 = pos
tw = th = size
if ow > tw or oh > th:
return cv2.crop((x1, y1, x1 + tw, y1 + th))
return img
@pytest.mark.parametrize(
"img_path",
[
"./test/Data/fish/fish_11.jpg",
],
)
def test_lambda(img_path):
pil_img = Image.open(img_path)
torchvision.set_image_backend("PIL")
torch.manual_seed(10)
pil_lambda = trans.Lambda(lambda img: custom_crop(pil_img, (5, 5), 224)),
torchvision.set_image_backend("cv2")
cv2_img = np.asarray(pil_img)
cv2_lambda = trans.Lambda(lambda img: custom_crop(cv2_img, (5, 5), 224)),
assert type(pil_lambda) == type(cv2_lambda)
pil_img.close()