def __getitem__(self, idx):
# img_name = self.root_dir + '/img_resized_1M/cities_instagram/' + self.indices[idx] + '.jpg'
if self.split == '/anns/validation':
img_name = '{}/{}/{}{}'.format(self.root_dir, 'img_val',
self.indices[idx], '.jpg')
else:
img_name = '{}/{}/{}{}'.format(self.root_dir, 'img',
self.indices[idx], '.jpg')
try:
image = Image.open(img_name)
# print("FOUND " + img_name)
except:
print("Img file " + img_name + " not found, using hardcoded " +
img_name)
img_name = '../../datasets/SocialMedia/img_resized_1M/cities_instagram/london/1481255189662056249.jpg'
image = Image.open(img_name)
try:
width, height = image.size
if self.RandomCrop >= width or self.RandomCrop >= height:
image = image.resize((int(width * 1.5), int(height * 1.5)),
Image.ANTIALIAS)
if self.Rescale != 0:
image = customTransform.Rescale(image, self.Rescale)
if self.RandomCrop != 0:
image = customTransform.RandomCrop(image, self.RandomCrop)
if self.Mirror:
image = customTransform.Mirror(image)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
except:
print("Error on data aumentation, using hardcoded")
img_name = self.root_dir + '/img_resized_1M/cities_instagram/london/1481255189662056249.jpg'
image = Image.open(img_name)
if self.RandomCrop != 0:
image = customTransform.RandomCrop(image, self.RandomCrop)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
out_img = np.copy(im_np)
# Simple Classification (class index)
# label = torch.from_numpy(np.array([int(self.labels[idx])]))
# label = label.type(torch.LongTensor)
# Multilabel / Regression
label = torch.from_numpy(np.array(self.labels[idx]))
label = label.type(torch.FloatTensor)
return torch.from_numpy(out_img), label
def __getitem__(self, idx):
img_name = '{}{}/{}{}'.format(self.root_dir, 'img_resized',
self.tweet_ids[idx], '.jpg') #, '.jpg')
try:
image = Image.open(img_name)
# print("FOUND " + img_name)
except:
new_img_name = '../../../datasets/HateSPic/MMHS/img_resized/1037385299310112768.jpg'
print("Img file " + img_name + " not found, using hardcoded " +
new_img_name)
image = Image.open(new_img_name)
try:
width, height = image.size
if self.RandomCrop >= width or self.RandomCrop >= height:
image = image.resize((int(width * 1.5), int(height * 1.5)),
Image.ANTIALIAS)
if self.Rescale != 0:
image = customTransform.Rescale(image, self.Rescale)
if self.RandomCrop != 0:
image = customTransform.RandomCrop(image, self.RandomCrop)
if self.Mirror:
image = customTransform.Mirror(image)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
except:
print("Error in data aumentation with image " + img_name)
img_name = '../../../datasets/HateSPic/MMHS/img_resized/1037385299310112768.jpg'
print("Using hardcoded: " + img_name)
image = Image.open(img_name)
if self.RandomCrop != 0:
image = customTransform.RandomCrop(image, self.RandomCrop)
if self.Rescale != 0:
image = customTransform.Rescale(image, self.Rescale)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
out_img = np.copy(im_np)
# Multilabel / Regression
img_text = torch.from_numpy(np.array(self.img_texts[idx]))
label = torch.from_numpy(np.array(self.labels[idx]))
tweet = torch.from_numpy(np.array(self.tweets[idx]))
# print(out_img.shape)
return torch.from_numpy(out_img), img_text, tweet, label
def __getitem__(self, idx):
if self.split == '/anns/validation':
img_name = '{}/{}/{}{}'.format(self.root_dir, 'img_val',
self.indices[idx], '.jpg')
else:
img_name = '{}/{}/{}{}'.format(self.root_dir, 'img',
self.indices[idx], '.jpg')
try:
image = Image.open(img_name)
except:
print("Img file " + img_name + " not found, using hardcoded " +
img_name)
img_name = '../../datasets/SocialMedia/img_resized_1M/cities_instagram/london/1481255189662056249.jpg'
image = Image.open(img_name)
if self.CenterCrop != 0:
crop_size = 256
image = customTransform.CenterCrop(image, crop_size,
self.CenterCrop)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
out_img = np.copy(im_np)
# Multilabel / Regression
label = torch.from_numpy(np.array(self.labels[idx]))
label = label.type(torch.FloatTensor)
return torch.from_numpy(out_img), label, self.indices[idx]
def __getitem__(self, idx):
img_name = '{}{}/{}{}'.format(self.root_dir, 'img_resized',
self.tweet_ids[idx], '.jpg')
try:
image = Image.open(img_name)
except:
new_img_name = '../../../datasets/HateSPic/MMHS/img_resized/1037385299310112768.jpg'
print("Img file " + img_name + " not found, using hardcoded " +
new_img_name)
image = Image.open(new_img_name)
try:
image = customTransform.Rescale(image, self.Rescale)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
except:
img_name = '../../../datasets/HateSPic/MMHS/img_resized/1037385299310112768.jpg'
print("Error on data aumentation, using hardcoded: " + img_name)
image = Image.open(img_name)
image = customTransform.Rescale(image, self.Rescale)
im_np = np.array(image, dtype=np.float32)
im_np = customTransform.PreprocessImage(im_np)
out_img = np.copy(im_np)
# Simple Classification (class index)
label = torch.from_numpy(np.array([int(self.labels[idx])]))
label = label.type(torch.LongTensor)
# Set text embedding to 0!
#self.img_texts[idx] = np.zeros(self.hidden_state_dim)
#self.tweets[idx] = np.zeros(self.hidden_state_dim)
# Set image to 0!
#out_img = np.zeros((3, 299, 299), dtype=np.float32)
# Multilabel / Regression
img_text = torch.from_numpy(np.array(self.img_texts[idx]))
tweet = torch.from_numpy(np.array(self.tweets[idx]))
# print(out_img.shape)
return self.tweet_ids[idx], torch.from_numpy(
out_img), img_text, tweet, label