97 lines
3.7 KiB
Python
97 lines
3.7 KiB
Python
import random
|
|
|
|
from PIL import Image
|
|
import PIL.Image
|
|
import numpy as np
|
|
|
|
|
|
def center_crop_arr(pil_image, image_size):
|
|
"""
|
|
Center cropping implementation from ADM.
|
|
https://github.com/openai/guided-diffusion/blob/8fb3ad9197f16bbc40620447b2742e13458d2831/guided_diffusion/image_datasets.py#L126
|
|
"""
|
|
while min(*pil_image.size) >= 2 * image_size:
|
|
pil_image = pil_image.resize(tuple(x // 2 for x in pil_image.size), resample=Image.BOX)
|
|
|
|
scale = image_size / min(*pil_image.size)
|
|
pil_image = pil_image.resize(tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC)
|
|
|
|
arr = np.array(pil_image)
|
|
crop_y = (arr.shape[0] - image_size) // 2
|
|
crop_x = (arr.shape[1] - image_size) // 2
|
|
return Image.fromarray(arr[crop_y : crop_y + image_size, crop_x : crop_x + image_size])
|
|
|
|
|
|
def center_crop(pil_image, crop_size):
|
|
while pil_image.size[0] >= 2 * crop_size[0] and pil_image.size[1] >= 2 * crop_size[1]:
|
|
pil_image = pil_image.resize(tuple(x // 2 for x in pil_image.size), resample=Image.BOX)
|
|
|
|
scale = max(crop_size[0] / pil_image.size[0], crop_size[1] / pil_image.size[1])
|
|
pil_image = pil_image.resize(tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC)
|
|
|
|
crop_left = random.randint(0, pil_image.size[0] - crop_size[0])
|
|
crop_upper = random.randint(0, pil_image.size[1] - crop_size[1])
|
|
crop_right = crop_left + crop_size[0]
|
|
crop_lower = crop_upper + crop_size[1]
|
|
return pil_image.crop(box=(crop_left, crop_upper, crop_right, crop_lower))
|
|
|
|
|
|
def pad(pil_image, pad_size):
|
|
while pil_image.size[0] >= 2 * pad_size[0] and pil_image.size[1] >= 2 * pad_size[1]:
|
|
pil_image = pil_image.resize(tuple(x // 2 for x in pil_image.size), resample=Image.BOX)
|
|
|
|
scale = min(pad_size[0] / pil_image.size[0], pad_size[1] / pil_image.size[1])
|
|
pil_image = pil_image.resize(tuple(round(x * scale) for x in pil_image.size), resample=Image.BICUBIC)
|
|
|
|
new_image = Image.new('RGB', pad_size, (255, 255, 255))
|
|
new_image.paste(pil_image, (0, 0))
|
|
return new_image
|
|
|
|
|
|
def var_center_crop(pil_image, crop_size_list, random_top_k=4):
|
|
w, h = pil_image.size
|
|
rem_percent = [min(cw / w, ch / h) / max(cw / w, ch / h) for cw, ch in crop_size_list]
|
|
crop_size = random.choice(
|
|
sorted(((x, y) for x, y in zip(rem_percent, crop_size_list)), reverse=True)[:random_top_k]
|
|
)[1]
|
|
return center_crop(pil_image, crop_size)
|
|
|
|
|
|
def var_pad(pil_image, pad_size_list, random_top_k=4):
|
|
w, h = pil_image.size
|
|
rem_percent = [min(cw / w, ch / h) / max(cw / w, ch / h) for cw, ch in pad_size_list]
|
|
crop_size = random.choice(
|
|
sorted(((x, y) for x, y in zip(rem_percent, pad_size_list)), reverse=True)[:random_top_k]
|
|
)[1]
|
|
return pad(pil_image, crop_size)
|
|
|
|
|
|
def match_size(w, h, crop_size_list, random_top_k=4):
|
|
rem_percent = [min(cw / w, ch / h) / max(cw / w, ch / h) for cw, ch in crop_size_list]
|
|
crop_size = random.choice(
|
|
sorted(((x, y) for x, y in zip(rem_percent, crop_size_list)), reverse=True)[:random_top_k]
|
|
)[1]
|
|
return crop_size
|
|
|
|
|
|
def generate_crop_size_list(num_patches, patch_size, max_ratio=4.0, step_size=1):
|
|
assert max_ratio >= 1.0
|
|
crop_size_list = []
|
|
wp, hp = num_patches, step_size
|
|
while wp > 0:
|
|
if max(wp, hp) / min(wp, hp) <= max_ratio:
|
|
crop_size_list.append((wp * patch_size, hp * patch_size))
|
|
if (hp + step_size) * wp <= num_patches:
|
|
hp += step_size
|
|
else:
|
|
wp -= step_size
|
|
return crop_size_list
|
|
|
|
def to_rgb_if_rgba(img: Image.Image):
|
|
if img.mode.upper() == "RGBA":
|
|
rgb_img = Image.new("RGB", img.size, (255, 255, 255))
|
|
rgb_img.paste(img, mask=img.split()[3]) # 3 is the alpha channel
|
|
return rgb_img
|
|
else:
|
|
return img
|