def load_dataset(opt):
"""Loads the input datasets."""
print('Reading dataset ', opt.dataset)
trainset = datasets.Fashion200k(
path=opt.dataset_path,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.Fashion200k(
path=opt.dataset_path,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
print('trainset size:', len(trainset))
print('testset size:', len(testset))
return trainset, testset
def GetValuestestloaded():
with open(Path1 + "\\BetatestLoaded.txt", 'rb') as fp:
BetaNormalize = pickle.load(fp)
trainset = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.Fashion200k(
path=Path1,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in trainset.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
opt = argparse.ArgumentParser()
opt.add_argument('--batch_size', type=int, default=2)
opt.add_argument('--dataset', type=str, default='fashion200k')
opt.batch_size = 1
opt.dataset = 'fashion200k'
for name, dataset in [('train', trainset),
('test', testset)]: #('train', trainset),
betaNor = test_retrieval.testLoadedBeta(opt, trig, dataset,
BetaNormalize)
print(name, ' BetaNormalized: ', betaNor)
asbook = test_retrieval.testLoaded(opt, trig, dataset)
print(name, ' As PaPer: ', asbook)
def ab_Mgetvaluesfilesaved(option):
trainset = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.Fashion200k(
path=Path1,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in trainset.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
opt = argparse.ArgumentParser()
opt.add_argument('--batch_size', type=int, default=2)
opt.add_argument('--dataset', type=str, default='fashion200k')
opt.batch_size = 1
opt.dataset = 'fashion200k'
#for name, dataset in [ ('train', trainset),('test', testset)]: #('train', trainset),
for name, dataset in [('test', testset)]:
#for name, dataset in [ ('train', trainset)]:
#('train', trainset),
asbook1, model, euc_model = ab_all_neural_testLoaded(
opt, trig, dataset, option)
print(name, ' Loaded As PaPer: ', asbook1, '\n model generated ',
model, '\n euc model', euc_model)
def testvaluessame():
train = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
transform = torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
])
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in train.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
trig.eval()
query = 'women/tops/blouses/91422080/91422080_0.jpeg'
qttext = 'replace sunrise with pleat-neck'
target = 'women/tops/sleeveless_and_tank_tops/90068628/90068628_0.jpeg'
text = []
text.append(qttext)
text.append(qttext)
img = Image.open(Path1 + '/' + query)
img = img.convert('RGB')
img = transform(img)
img2 = Image.open(Path1 + '/' + target)
img2 = img2.convert('RGB')
img2 = transform(img2)
img = img.unsqueeze_(0)
img2 = img2.unsqueeze_(0)
images = torch.cat([img, img2], dim=0)
trigdataQ = trig.compose_img_text(images, text)
trigdataQ1 = trig.compose_img_text(images, text)
print('...........')
print(trigdataQ)
print(trigdataQ1)
def savesourcevalues():
train = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
test = datasets.Fashion200k(
path=Path1,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in train.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
opt = argparse.ArgumentParser()
opt.add_argument('--batch_size', type=int, default=2)
opt.add_argument('--dataset', type=str, default='fashion200k')
opt.batch_size = 1
opt.dataset = 'fashion200k'
datasets.Features172K().SavetoFilesImageSource(Path1 + r'/dataset172',
trig, train, opt)
print('172 Finished')
def load_dataset(opt):
"""Loads the input datasets."""
print('Reading dataset: ', opt.dataset)
if opt.dataset == 'fashion200k':
trainset = datasets.Fashion200k(
path=opt.dataset_path,
split='train',
transform=torchvision.transforms.Compose([
# Resize 的输入为一个数的时候，会将维度最低的转换为输入的数，同时图像等比例缩放
# 也可以输入tuple，即转换为tuple大小
torchvision.transforms.Resize(224),
# CenterCrop(x) 即从中心位置裁剪一个输入大小 x 的图像
torchvision.transforms.CenterCrop(224),
# ToTensor会将PIL类型的图片(H, W, C) 转换为 (C, H, W)，同时除以255，即所有元素都在[0,1]
# 如果要将(C, H, W) 转换回 (H, W, C) 可以使用tensor的permute(1, 2, 0)
torchvision.transforms.ToTensor(),
# 此处输入的参数，第一个列表分别是3个通道的mean(均值)
# 第二个列表分别是3个通道的std(方差)
# Normalize --> output[channel] = (input[channel] - mean[channel]) / std[channel]
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225]),
]))
testset = datasets.Fashion200k(
path=opt.dataset_path,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
else:
print('Invalid dataset: ', opt.dataset)
sys.exit()
print('trainset size: ', str(len(trainset)))
print('testset size: ', str(len(testset)))
return trainset, testset
def load_dataset(opt):
"""Loads the input datasets."""
print('Reading dataset ', opt.dataset)
if opt.dataset == 'fashion200k':
trainset = datasets.Fashion200k(
path=opt.dataset_path,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.Fashion200k(
path=opt.dataset_path,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
elif opt.dataset == 'mitstates':
trainset = datasets.MITStates(
path=opt.dataset_path,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.MITStates(
path=opt.dataset_path,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
elif opt.dataset == 'fashionIQ':
trainset = datasets.FashionIQ(
path=opt.dataset_path,
cat_type=opt.category_to_train,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
testset = datasets.FashionIQ(
path=opt.dataset_path,
cat_type=opt.category_to_train,
split='val',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
else:
print('Invalid dataset', opt.dataset)
sys.exit()
print('trainset size:', len(trainset))
print('testset size:', len(testset))
return trainset, testset
def distanceBetaand():
with open(Path1 + "/Beta.txt", 'rb') as fp:
Beta = pickle.load(fp)
trainset = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
test = datasets.Fashion200k(
path=Path1,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in trainset.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
trig.eval()
imgs = []
mods = []
target = []
batchsize = 2
Distance = []
sourceid = []
targetid = []
countbeta = 0
counttrig = 0
for Data in tqdm(trainset):
imgs += [Data['source_img_data']]
mods += [Data['mod']['str']]
target += [Data['target_img_data']]
sourceid.append(Data['source_img_id'])
targetid.append(Data['target_img_id'])
imgs = torch.stack(imgs).float()
imgs = torch.autograd.Variable(imgs)
f = trig.compose_img_text(imgs, mods).data.cpu().numpy()
target = torch.stack(target).float()
target = torch.autograd.Variable(target)
f2 = trig.extract_img_feature(target).data.cpu().numpy()
trigdata = f[0]
trigbeta = np.insert(trigdata, 0, 1)
trigbeta = np.matmul(trigbeta, Beta)
Targetdata = f2[0]
SourceTarget = euclideandistance(trigdata, Targetdata)
betaTarget = euclideandistance(trigbeta, Targetdata)
if (SourceTarget > betaTarget):
countbeta = countbeta + 1
else:
counttrig = counttrig + 1
# opsig={'source':sourceid[0],'target':targetid[0],'disbeta':betaTarget,'disorig':SourceTarget}
# Distance.append(opsig )
imgs = []
mods = []
target = []
sourceid = []
targetid = []
with open(Path1 + r"/" + 'Distance.txt', 'wb') as fp:
pickle.dump(Distance, fp)
print('Train Data :Count beta less:', countbeta, ' ,countbeta bigger:',
counttrig)
imgs = []
mods = []
target = []
batchsize = 2
Distance = []
sourceid = []
targetid = []
countbeta = 0
counttrig = 0
for Data in tqdm(test.get_test_queries()):
imgs += [test.get_img(Data['source_img_id'])]
mods += [Data['mod']['str']]
target += [test.get_img(Data['target_id'])]
imgs = torch.stack(imgs).float()
imgs = torch.autograd.Variable(imgs)
f = trig.compose_img_text(imgs, mods).data.cpu().numpy()
target = torch.stack(target).float()
target = torch.autograd.Variable(target)
f2 = trig.extract_img_feature(target).data.cpu().numpy()
trigdata = f[0]
trigbeta = np.insert(trigdata, 0, 1)
trigbeta = np.matmul(trigbeta, Beta)
Targetdata = f2[0]
SourceTarget = euclideandistance(trigdata, Targetdata)
betaTarget = euclideandistance(trigbeta, Targetdata)
if (SourceTarget > betaTarget):
countbeta = countbeta + 1
else:
counttrig = counttrig + 1
imgs = []
mods = []
target = []
sourceid = []
targetid = []
print('Test Data :Count beta less:', countbeta, ' ,countbeta bigger:',
counttrig)
def GetValuestrain15time():
with open(Path1 + "/trainBetaNormalized.txt", 'rb') as fp:
BetaNormalize = pickle.load(fp)
trainset = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trainloader = trainset.get_loader(batch_size=2,
shuffle=True,
drop_last=True,
num_workers=0)
testset = TestFashion200k(
path=Path1,
split='test',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = TIRG([t.encode().decode('utf-8') for t in trainset.get_all_texts()],
512)
trig.load_state_dict(
torch.load(Path1 + r'\checkpoint_fashion200k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
opt = argparse.ArgumentParser()
opt.add_argument('--batch_size', type=int, default=2)
opt.add_argument('--dataset', type=str, default='fashion200k')
opt.batch_size = 1
opt.dataset = 'fashion200k'
Results = []
for i in range(15):
for name, dataset in [('train', trainset)]: #,('test', testset)]:
# betaNor="['1 ---> 5.27', '5 ---> 14.39', '10 ---> 21.6', '50 ---> 43.830000000000005', '100 ---> 55.33']"
# Results.append('No.'+str(i)+' DataSet='+name+' Type= BetaNormalized '+' Result=' +betaNor)
try:
betaNor = test_retrieval.testbetanormalizednot(
opt, trig, dataset, BetaNormalize)
print(name, ' BetaNormalized: ', betaNor)
Results.append('No.' + str(i) + ' DataSet=' + name +
' Type= BetaNormalized ' + ' Result=' + betaNor)
except:
print('ERROR')
try:
asbook = test_retrieval.test(opt, trig, dataset)
print(name, ' As PaPer: ', asbook)
Results.append('No.' + str(i) + ' DataSet=' + name +
' Type= As PaPer ' + ' Result=' + betaNor)
except:
print('ERROR')
with open(Path1 + r"/" + 'Results15time.txt', 'wb') as fp:
pickle.dump(Results, fp)
def getbetatrainNot():
train = datasets.Fashion200k(
path=Path1,
split='train',
transform=torchvision.transforms.Compose([
torchvision.transforms.Resize(224),
torchvision.transforms.CenterCrop(224),
torchvision.transforms.ToTensor(),
torchvision.transforms.Normalize([0.485, 0.456, 0.406],
[0.229, 0.224, 0.225])
]))
trig = img_text_composition_models.TIRG(
[t.encode().decode('utf-8') for t in train.get_all_texts()], 512)
trig.load_state_dict(
torch.load(Path1 + r'\fashion200k.tirg.iter160k.pth',
map_location=torch.device('cpu'))['model_state_dict'])
trig.eval()
imgs = []
mods = []
trigdata = []
target = []
imgdata = []
for Data in tqdm(train):
imgs += [Data['source_img_data']]
mods += [Data['mod']['str']]
target += [Data['target_img_data']]
imgs = torch.stack(imgs).float()
imgs = torch.autograd.Variable(imgs)
f = trig.compose_img_text(imgs, mods).data.cpu().numpy()
target = torch.stack(target).float()
target = torch.autograd.Variable(target)
f2 = trig.extract_img_feature(target).data.cpu().numpy()
trigdata.append(f[0])
imgdata.append(f2[0])
imgs = []
mods = []
target = []
trigdata = np.array(trigdata)
imgdata = np.array(imgdata)
Ntrigdata = trigdata
Nimgdata = imgdata
Ntrig2 = []
for i in range(Ntrigdata.shape[0]):
Ntrigdata[i, :] /= np.linalg.norm(Ntrigdata[i, :])
for i in range(Nimgdata.shape[0]):
Nimgdata[i, :] /= np.linalg.norm(Nimgdata[i, :])
for i in range(Ntrigdata.shape[0]):
Ntrig2.append(np.insert(Ntrigdata[i], 0, 1))
Ntrig2 = np.array(Ntrig2)
Ntrigdata1 = Ntrig2.transpose()
X1 = np.matmul(Ntrigdata1, Ntrig2)
X2 = np.linalg.inv(X1)
X3 = np.matmul(X2, Ntrigdata1)
Nbeta = np.matmul(X3, Nimgdata)
with open(Path1 + r"/" + 'BetaNot.txt', 'wb') as fp:
pickle.dump(Nbeta, fp)
