EveryDream-trainer/ldm/data/image_train_item.py

130 lines
5.1 KiB
Python

import PIL
import numpy as np
from torchvision import transforms, utils
import random
import math
import os
_RANDOM_TRIM = 0.04
class ImageTrainItem():
"""
image: PIL.Image
identifier: caption,
target_aspect: (width, height),
pathname: path to image file
flip_p: probability of flipping image (0.0 to 1.0)
"""
def __init__(self, image: PIL.Image, caption: str, target_wh: list, pathname: str, flip_p=0.0):
self.caption = caption
self.target_wh = target_wh
self.pathname = pathname
self.flip = transforms.RandomHorizontalFlip(p=flip_p)
self.cropped_img = None
if image is None:
self.image = []
else:
self.image = image
def hydrate(self, crop=False, save=False, crop_jitter=20):
"""
crop: hard center crop to 512x512
save: save the cropped image to disk, for manual inspection of resize/crop
crop_jitter: randomly shift cropp by N pixels when using multiple aspect ratios to improve training quality
"""
if not hasattr(self, 'image') or len(self.image) == 0:
self.image = PIL.Image.open(self.pathname).convert('RGB')
width, height = self.image.size
if crop:
cropped_img = self.__autocrop(self.image)
self.image = cropped_img.resize((512,512), resample=PIL.Image.BICUBIC)
else:
width, height = self.image.size
jitter_amount = random.randint(0,crop_jitter)
if self.target_wh[0] == self.target_wh[1]:
if width > height:
left = random.randint(0, width - height)
self.image = self.image.crop((left, 0, height+left, height))
width = height
elif height > width:
top = random.randint(0, height - width)
self.image = self.image.crop((0, top, width, width+top))
height = width
elif width > self.target_wh[0]:
slice = min(int(self.target_wh[0] * _RANDOM_TRIM), width-self.target_wh[0])
slicew_ratio = random.random()
left = int(slice*slicew_ratio)
right = width-int(slice*(1-slicew_ratio))
sliceh_ratio = random.random()
top = int(slice*sliceh_ratio)
bottom = height- int(slice*(1-sliceh_ratio))
self.image = self.image.crop((left, top, right, bottom))
else:
image_aspect = width / height
target_aspect = self.target_wh[0] / self.target_wh[1]
if image_aspect > target_aspect:
new_width = int(height * target_aspect)
jitter_amount = max(min(jitter_amount, int(abs(width-new_width)/2)), 0)
left = jitter_amount
right = left + new_width
self.image = self.image.crop((left, 0, right, height))
else:
new_height = int(width / target_aspect)
jitter_amount = max(min(jitter_amount, int(abs(height-new_height)/2)), 0)
top = jitter_amount
bottom = top + new_height
self.image = self.image.crop((0, top, width, bottom))
self.image = self.image.resize(self.target_wh, resample=PIL.Image.BICUBIC)
self.image = self.flip(self.image)
if type(self.image) is not np.ndarray:
if save:
base_name = os.path.basename(self.pathname)
if not os.path.exists("test/output"):
os.makedirs("test/output")
self.image.save(f"test/output/{base_name}")
self.image = np.array(self.image).astype(np.uint8)
self.image = (self.image / 127.5 - 1.0).astype(np.float32)
#print(self.image.shape)
return self
@staticmethod
def __autocrop(image: PIL.Image, q=.404):
"""
crops image to a random square inside small axis using a truncated gaussian distribution across the long axis
"""
x, y = image.size
if x != y:
if (x>y):
rand_x = x-y
sigma = max(rand_x*q,1)
else:
rand_y = y-x
sigma = max(rand_y*q,1)
if (x>y):
x_crop_gauss = abs(random.gauss(0, sigma))
x_crop = min(x_crop_gauss,(x-y)/2)
x_crop = math.trunc(x_crop)
y_crop = 0
else:
y_crop_gauss = abs(random.gauss(0, sigma))
x_crop = 0
y_crop = min(y_crop_gauss,(y-x)/2)
y_crop = math.trunc(y_crop)
min_xy = min(x, y)
image = image.crop((x_crop, y_crop, x_crop + min_xy, y_crop + min_xy))
return image