Spaces:
Running
Running
File size: 7,607 Bytes
13580fb |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 |
# Copyright Amazon.com, Inc. or its affiliates. All Rights Reserved.
# SPDX-License-Identifier: MIT
import random
import torch
from torch.utils.data import Dataset
from torch.utils.data import sampler
#import lmdb
import torchvision.transforms as transforms
import six
import sys
from PIL import Image
import numpy as np
import os
import sys
import pickle
import numpy as np
from params import *
import glob, cv2
import torchvision.transforms as transforms
def crop_(input):
image = Image.fromarray(input)
image = image.convert('L')
binary_image = image.point(lambda x: 0 if x > 127 else 255, '1')
bbox = binary_image.getbbox()
cropped_image = image.crop(bbox)
return np.array(cropped_image)
def get_transform(grayscale=False, convert=True):
transform_list = []
if grayscale:
transform_list.append(transforms.Grayscale(1))
if convert:
transform_list += [transforms.ToTensor()]
if grayscale:
transform_list += [transforms.Normalize((0.5,), (0.5,))]
else:
transform_list += [transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]
return transforms.Compose(transform_list)
def load_itw_samples(folder_path, num_samples = 15):
if isinstance(folder_path, str):
paths = glob.glob(f'{folder_path}/*')
else:
paths = folder_path
paths = np.random.choice(paths, num_samples, replace = len(paths)<=num_samples)
words = [os.path.basename(path_i)[:-4] for path_i in paths]
imgs = [np.array(Image.open(i).convert('L')) for i in paths]
imgs = [crop_(im) for im in imgs]
imgs = [cv2.resize(imgs_i, (int(32*(imgs_i.shape[1]/imgs_i.shape[0])), 32)) for imgs_i in imgs]
max_width = 192
imgs_pad = []
imgs_wids = []
trans_fn = get_transform(grayscale=True)
for img in imgs:
img = 255 - img
img_height, img_width = img.shape[0], img.shape[1]
outImg = np.zeros(( img_height, max_width), dtype='float32')
outImg[:, :img_width] = img[:, :max_width]
img = 255 - outImg
imgs_pad.append(trans_fn((Image.fromarray(img))))
imgs_wids.append(img_width)
imgs_pad = torch.cat(imgs_pad, 0)
return imgs_pad.unsqueeze(0), torch.Tensor(imgs_wids).unsqueeze(0)
class TextDataset():
def __init__(self, base_path = DATASET_PATHS, num_examples = 15, target_transform=None):
self.NUM_EXAMPLES = num_examples
#base_path = DATASET_PATHS
file_to_store = open(base_path, "rb")
self.IMG_DATA = pickle.load(file_to_store)['train']
self.IMG_DATA = dict(list( self.IMG_DATA.items())) #[:NUM_WRITERS])
if 'None' in self.IMG_DATA.keys():
del self.IMG_DATA['None']
self.author_id = list(self.IMG_DATA.keys())
self.transform = get_transform(grayscale=True)
self.target_transform = target_transform
self.collate_fn = TextCollator()
def __len__(self):
return len(self.author_id)
def __getitem__(self, index):
NUM_SAMPLES = self.NUM_EXAMPLES
author_id = self.author_id[index]
self.IMG_DATA_AUTHOR = self.IMG_DATA[author_id]
random_idxs = np.random.choice(len(self.IMG_DATA_AUTHOR), NUM_SAMPLES, replace = True)
rand_id_real = np.random.choice(len(self.IMG_DATA_AUTHOR))
real_img = self.transform(self.IMG_DATA_AUTHOR[rand_id_real]['img'].convert('L'))
real_labels = self.IMG_DATA_AUTHOR[rand_id_real]['label'].encode()
imgs = [np.array(self.IMG_DATA_AUTHOR[idx]['img'].convert('L')) for idx in random_idxs]
labels = [self.IMG_DATA_AUTHOR[idx]['label'].encode() for idx in random_idxs]
max_width = 192 #[img.shape[1] for img in imgs]
imgs_pad = []
imgs_wids = []
for img in imgs:
img = 255 - img
img_height, img_width = img.shape[0], img.shape[1]
outImg = np.zeros(( img_height, max_width), dtype='float32')
outImg[:, :img_width] = img[:, :max_width]
img = 255 - outImg
imgs_pad.append(self.transform((Image.fromarray(img))))
imgs_wids.append(img_width)
imgs_pad = torch.cat(imgs_pad, 0)
item = {'simg': imgs_pad, 'swids':imgs_wids, 'img' : real_img, 'label':real_labels,'img_path':'img_path', 'idx':'indexes', 'wcl':index}
return item
class TextDatasetval():
def __init__(self, base_path = DATASET_PATHS, num_examples = 15, target_transform=None):
self.NUM_EXAMPLES = num_examples
#base_path = DATASET_PATHS
file_to_store = open(base_path, "rb")
self.IMG_DATA = pickle.load(file_to_store)['test']
self.IMG_DATA = dict(list( self.IMG_DATA.items()))#[NUM_WRITERS:])
if 'None' in self.IMG_DATA.keys():
del self.IMG_DATA['None']
self.author_id = list(self.IMG_DATA.keys())
self.transform = get_transform(grayscale=True)
self.target_transform = target_transform
self.collate_fn = TextCollator()
def __len__(self):
return len(self.author_id)
def __getitem__(self, index):
NUM_SAMPLES = self.NUM_EXAMPLES
author_id = self.author_id[index]
self.IMG_DATA_AUTHOR = self.IMG_DATA[author_id]
random_idxs = np.random.choice(len(self.IMG_DATA_AUTHOR), NUM_SAMPLES, replace = True)
rand_id_real = np.random.choice(len(self.IMG_DATA_AUTHOR))
real_img = self.transform(self.IMG_DATA_AUTHOR[rand_id_real]['img'].convert('L'))
real_labels = self.IMG_DATA_AUTHOR[rand_id_real]['label'].encode()
imgs = [np.array(self.IMG_DATA_AUTHOR[idx]['img'].convert('L')) for idx in random_idxs]
labels = [self.IMG_DATA_AUTHOR[idx]['label'].encode() for idx in random_idxs]
max_width = 192 #[img.shape[1] for img in imgs]
imgs_pad = []
imgs_wids = []
for img in imgs:
img = 255 - img
img_height, img_width = img.shape[0], img.shape[1]
outImg = np.zeros(( img_height, max_width), dtype='float32')
outImg[:, :img_width] = img[:, :max_width]
img = 255 - outImg
imgs_pad.append(self.transform((Image.fromarray(img))))
imgs_wids.append(img_width)
imgs_pad = torch.cat(imgs_pad, 0)
item = {'simg': imgs_pad, 'swids':imgs_wids, 'img' : real_img, 'label':real_labels,'img_path':'img_path', 'idx':'indexes', 'wcl':index}
return item
class TextCollator(object):
def __init__(self):
self.resolution = resolution
def __call__(self, batch):
img_path = [item['img_path'] for item in batch]
width = [item['img'].shape[2] for item in batch]
indexes = [item['idx'] for item in batch]
simgs = torch.stack([item['simg'] for item in batch], 0)
wcls = torch.Tensor([item['wcl'] for item in batch])
swids = torch.Tensor([item['swids'] for item in batch])
imgs = torch.ones([len(batch), batch[0]['img'].shape[0], batch[0]['img'].shape[1], max(width)], dtype=torch.float32)
for idx, item in enumerate(batch):
try:
imgs[idx, :, :, 0:item['img'].shape[2]] = item['img']
except:
print(imgs.shape)
item = {'img': imgs, 'img_path':img_path, 'idx':indexes, 'simg': simgs, 'swids': swids, 'wcl':wcls}
if 'label' in batch[0].keys():
labels = [item['label'] for item in batch]
item['label'] = labels
if 'z' in batch[0].keys():
z = torch.stack([item['z'] for item in batch])
item['z'] = z
return item
|