Here:
|
img299 = cv2.resize(img224, (299, 299)) |
This resizing is a bilinear interpolation, which can produce some loss. Nearest neighbor is said to be lossless:
resized_image = cv2.resize(image, (new_width, new_height), interpolation=cv2.INTER_NEAREST)
Here:
Beyond-ImageNet-Attack/imagenet/convert.py
Line 22 in 863b758
This resizing is a bilinear interpolation, which can produce some loss. Nearest neighbor is said to be lossless: