Spaces:
Sleeping
Sleeping
File size: 4,112 Bytes
9c4b1c4 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 |
import cv2
import numpy as np
import torchvision.datasets as datasets
import torchvision.transforms as transforms
import torchvision.transforms.functional as TF
from random import random, choice
from io import BytesIO
from PIL import Image
from PIL import ImageFile
from scipy.ndimage.filters import gaussian_filter
from torchvision.transforms import InterpolationMode
ImageFile.LOAD_TRUNCATED_IMAGES = True
def dataset_folder(opt, root):
if opt.mode == 'binary':
return binary_dataset(opt, root)
if opt.mode == 'filename':
return FileNameDataset(opt, root)
raise ValueError('opt.mode needs to be binary or filename.')
def binary_dataset(opt, root):
if opt.isTrain:
crop_func = transforms.RandomCrop(opt.cropSize)
elif opt.no_crop:
crop_func = transforms.Lambda(lambda img: img)
else:
crop_func = transforms.CenterCrop(opt.cropSize)
if opt.isTrain and not opt.no_flip:
flip_func = transforms.RandomHorizontalFlip()
else:
flip_func = transforms.Lambda(lambda img: img)
if not opt.isTrain and opt.no_resize:
rz_func = transforms.Lambda(lambda img: img)
else:
# rz_func = transforms.Lambda(lambda img: custom_resize(img, opt))
rz_func = transforms.Resize((opt.loadSize, opt.loadSize))
dset = datasets.ImageFolder(
root,
transforms.Compose([
rz_func,
# transforms.Lambda(lambda img: data_augment(img, opt)),
crop_func,
flip_func,
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
]))
return dset
class FileNameDataset(datasets.ImageFolder):
def name(self):
return 'FileNameDataset'
def __init__(self, opt, root):
self.opt = opt
super().__init__(root)
def __getitem__(self, index):
# Loading sample
path, target = self.samples[index]
return path
def data_augment(img, opt):
img = np.array(img)
if random() < opt.blur_prob:
sig = sample_continuous(opt.blur_sig)
gaussian_blur(img, sig)
if random() < opt.jpg_prob:
method = sample_discrete(opt.jpg_method)
qual = sample_discrete(opt.jpg_qual)
img = jpeg_from_key(img, qual, method)
return Image.fromarray(img)
def sample_continuous(s):
if len(s) == 1:
return s[0]
if len(s) == 2:
rg = s[1] - s[0]
return random() * rg + s[0]
raise ValueError("Length of iterable s should be 1 or 2.")
def sample_discrete(s):
if len(s) == 1:
return s[0]
return choice(s)
def gaussian_blur(img, sigma):
gaussian_filter(img[:,:,0], output=img[:,:,0], sigma=sigma)
gaussian_filter(img[:,:,1], output=img[:,:,1], sigma=sigma)
gaussian_filter(img[:,:,2], output=img[:,:,2], sigma=sigma)
def cv2_jpg(img, compress_val):
img_cv2 = img[:,:,::-1]
encode_param = [int(cv2.IMWRITE_JPEG_QUALITY), compress_val]
result, encimg = cv2.imencode('.jpg', img_cv2, encode_param)
decimg = cv2.imdecode(encimg, 1)
return decimg[:,:,::-1]
def pil_jpg(img, compress_val):
out = BytesIO()
img = Image.fromarray(img)
img.save(out, format='jpeg', quality=compress_val)
img = Image.open(out)
# load from memory before ByteIO closes
img = np.array(img)
out.close()
return img
jpeg_dict = {'cv2': cv2_jpg, 'pil': pil_jpg}
def jpeg_from_key(img, compress_val, key):
method = jpeg_dict[key]
return method(img, compress_val)
# rz_dict = {'bilinear': Image.BILINEAR,
# 'bicubic': Image.BICUBIC,
# 'lanczos': Image.LANCZOS,
# 'nearest': Image.NEAREST}
rz_dict = {'bilinear': InterpolationMode.BILINEAR,
'bicubic': InterpolationMode.BICUBIC,
'lanczos': InterpolationMode.LANCZOS,
'nearest': InterpolationMode.NEAREST}
def custom_resize(img, opt):
interp = sample_discrete(opt.rz_interp)
return TF.resize(img, (opt.loadSize,opt.loadSize), interpolation=rz_dict[interp])
|