def curet_Train_test():
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ["sample01", "sample02", "sample03", "sample04", "sample05", "sample05", "sample06",
"sample07",
"sample08", "sample09", "sample10", "sample11", "sample12", "sample13", "sample14",
"sample15",
"sample16", "sample17", "sample18", "sample19", "sample20", "sample21", "sample22",
"sample23",
"sample24", "sample25", "sample26", "sample27", "sample28", "sample29", "sample30",
"sample31",
"sample32", "sample33", "sample34", "sample35", "sample36", "sample37", "sample38",
"sample39",
"sample40", "sample41", "sample42", "sample43", "sample44", "sample45", "sample46",
"sample47",
"sample48", "sample49", "sample50", "sample51", "sample52", "sample53", "sample54",
"sample55",
"sample56", "sample57", "sample58", "sample59", "sample60", "sample61"]
IMAGE_NET_LABELS = \
["alp", "artichoke", "Band Aid", "bathing cap", "bookshop", "bull mastiff", "butcher shop",
"carbonara", "chain", "chain saw", "chainlink fence", "cheetah", "cliff dwelling", "common iguana",
"confectionery", "container ship", "corn", "crossword puzzle", "dishrag", "dock", "flat-coated retriever",
"gibbon", "grocery store", "head cabbage", "honeycomb", "hook", "hourglass", "jigsaw puzzle",
"jinrikisha", "lakeside", "lawn mower", "maillot", "microwave", "miniature poodle", "muzzle",
"notebook", "ocarina", "orangutan", "organ", "paper towel", "partridge", "rapeseed",
"sandbar", "sarong", "sea urchin", "shoe shop", "shower curtain", "stone wall", "theater curtain",
"tile roof",
"turnstile", "vault", "velvet", "window screen", "wool", "yellow lady's slipper"]
train_parameters = {
"epochs": 75,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
image_net_data_set_path = "./Dataset/ImageNet/ImageNet_X.pickle"
image_net_label_set_path = "./Dataset/ImageNet/ImageNet_Y.pickle"
image_net_test_path = "./Dataset/ImageNet/ImageNet_Test.pickle"
texture_train_data_set_path = "./Dataset/Texture/CURET/CURET_X.pickle"
texture_train_label_set_path = "./Dataset/Texture/CURET/CURET_Y.pickle"
saved_model_name = "./Models/MTL/CURET/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.20
# training starts
texture_train_val_data_loader_list, texture_test_data_loader_list = \
DataPreProcessor.preprocess_texture_except_DTD(texture_train_data_set_path,
texture_train_label_set_path,
train_parameters[
"texture_batch_size"],
num_workers=0,
device=device,
split_size=split_size,
type="CURET")
image_net_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_data_set_path,
image_net_label_set_path,
train_parameters[
"image_net_batch_size"],
image_net_test_path,
num_workers=0,
split_size=split_size,
device=device,
type="ImageNet")
train_arguments = {
"IMAGE_NET_LABELS": IMAGE_NET_LABELS,
"TEXTURE_LABELS": TEXTURE_LABELS,
"image_net_data_loader_dict": image_net_data_loader_dict,
"texture_data_loader_list": texture_train_val_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
network = train.train_classifier(train_arguments, device, dataset_name="CURET")
# training ends
# testing starts
model_path_bn = "./Models/MTL/CURET/Multitask_Classifier_Model_epoch_75_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": texture_test_data_loader_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
test = Test_Classifier()
accuracy_list, mean_accuracy = test.test_classifier(test_arguments, IMAGE_NET_LABELS, device,
dataset_name="CURET")
file1 = open("CURET_Details.txt", "a")
file1.write(str(train_parameters))
file1.write("CURET Mean accuracy: {0}\n".format(mean_accuracy))
file1.write(str(accuracy_list))
pd.DataFrame.from_dict(
accuracy_list,
orient='columns'
).to_csv("./Accuracy_CURET.csv")
def kylberg_Train_test():
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ["Kyberge_blanket1", "Kyberge_blanket2", "Kyberge_canvas1",
"Kyberge_ceiling1", "Kyberge_ceiling2", "Kyberge_cushion1",
"Kyberge_floor1", "Kyberge_floor2", "Kyberge_grass1", "Kyberge_lentils1",
"Kyberge_linseeds1", "Kyberge_oatmeal1", "Kyberge_pearlsugar1", "Kyberge_rice1",
"Kyberge_rice2", "Kyberge_rug1", "Kyberge_sand1", "Kyberge_scarf1", "Kyberge_scarf2",
"Kyberge_sesameseeds1", "Kyberge_stone1", "Kyberge_stone2", "Kyberge_stone3",
"Kyberge_stoneslab1", "Kyberge_wall1"]
IMAGE_NET_LABELS = \
["alp", "artichoke", "Band Aid", "bathing cap", "bookshop", "bull mastiff", "butcher shop",
"carbonara", "chain", "chain saw", "chainlink fence", "cheetah", "cliff dwelling", "common iguana",
"confectionery", "container ship", "corn", "crossword puzzle", "dishrag", "dock", "flat-coated retriever",
"gibbon", "grocery store", "head cabbage", "honeycomb", "hook", "hourglass", "jigsaw puzzle",
"jinrikisha", "lakeside", "lawn mower", "maillot", "microwave", "miniature poodle", "muzzle",
"notebook", "ocarina", "orangutan", "organ", "paper towel", "partridge", "rapeseed",
"sandbar", "sarong", "sea urchin", "shoe shop", "shower curtain", "stone wall", "theater curtain",
"tile roof",
"turnstile", "vault", "velvet", "window screen", "wool", "yellow lady's slipper"]
train_parameters = {
"epochs": 400,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
image_net_data_set_path = "./Dataset/ImageNet/ImageNet_X.pickle"
image_net_label_set_path = "./Dataset/ImageNet/ImageNet_Y.pickle"
image_net_test_path = "./Dataset/ImageNet/ImageNet_Test.pickle"
texture_train_data_set_path = "./Dataset/Texture/kylberg/kylbergs_X.pickle"
texture_train_label_set_path = "./Dataset/Texture/kylberg/kylbergs_Y.pickle"
saved_model_name = "./Models/MTL/kylberg/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.20
_accuracy_list = []
# training starts
texture_train_val_data_loader_list, texture_test_data_loader_list = \
DataPreProcessor.preprocess_texture_except_DTD(texture_train_data_set_path,
texture_train_label_set_path,
train_parameters[
"texture_batch_size"],
num_workers=0,
device=device,
split_size=split_size,
type="kylberg",
folds=1)
image_net_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_data_set_path,
image_net_label_set_path,
train_parameters[
"image_net_batch_size"],
image_net_test_path,
num_workers=0,
split_size=split_size,
device=device,
type="ImageNet")
train_arguments = {
"IMAGE_NET_LABELS": IMAGE_NET_LABELS,
"TEXTURE_LABELS": TEXTURE_LABELS,
"image_net_data_loader_dict": image_net_data_loader_dict,
"texture_data_loader_list": texture_train_val_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
network = train.train_classifier(train_arguments, device, dataset_name="kylberg")
# training ends
# testing starts
model_path_bn = "./Models/MTL/kylberg/Multitask_Classifier_Model_epoch_400_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": texture_test_data_loader_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
test = Test_Classifier()
accuracy_list, mean_accuracy = test.test_classifier(test_arguments, IMAGE_NET_LABELS, device,
dataset_name="kylberg")
file1 = open("kylberg_Details.txt", "a")
file1.write(str(train_parameters))
file1.write("kylberg Mean accuracy: {0}\n".format(mean_accuracy))
file1.write(str(accuracy_list))
pd.DataFrame.from_dict(
accuracy_list,
orient='columns'
).to_csv("./Accuracy_kylberg.csv")
def uiuc_Train_test():
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ["UIUC01_bark1", "UIUC02_bark2", "UIUC03_bark3", "UIUC04_wood1", "UIUC05_wood2",
"UIUC06_wood3",
"UIUC07_water", "UIUC08_granite", "UIUC09_marble", "UIUC10_floor1", "UIUC11_floor2",
"UIUC12_pebbles", "UIUC13_wall", "UIUC14_brick1", "UIUC15_brick2", "UIUC16_glass1",
"UIUC17_glass2",
"UIUC18_carpet1", "UIUC19_carpet2", "UIUC20_upholstery", "UIUC21_wallpaper", "UIUC22_fur",
"UIUC23_knit", "UIUC24_corduroy", "UIUC25_plaid"]
IMAGE_NET_LABELS = \
["alp", "artichoke", "Band Aid", "bathing cap", "bookshop", "bull mastiff", "butcher shop",
"carbonara", "chain", "chain saw", "chainlink fence", "cheetah", "cliff dwelling", "common iguana",
"confectionery", "container ship", "corn", "crossword puzzle", "dishrag", "dock", "flat-coated retriever",
"gibbon", "grocery store", "head cabbage", "honeycomb", "hook", "hourglass", "jigsaw puzzle",
"jinrikisha", "lakeside", "lawn mower", "maillot", "microwave", "miniature poodle", "muzzle",
"notebook", "ocarina", "orangutan", "organ", "paper towel", "partridge", "rapeseed",
"sandbar", "sarong", "sea urchin", "shoe shop", "shower curtain", "stone wall", "theater curtain",
"tile roof",
"turnstile", "vault", "velvet", "window screen", "wool", "yellow lady's slipper"]
train_parameters = {
"epochs": 75,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
image_net_data_set_path = "./Dataset/ImageNet/ImageNet_X.pickle"
image_net_label_set_path = "./Dataset/ImageNet/ImageNet_Y.pickle"
image_net_test_path = "./Dataset/ImageNet/ImageNet_Test.pickle"
texture_train_data_set_path = "./Dataset/Texture/UIUC/UIUC_X.pickle"
texture_train_label_set_path = "./Dataset/Texture/UIUC/UIUC_Y.pickle"
saved_model_name = "./Models/MTL/UIUC/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.20
# training starts
texture_train_val_data_loader_list, texture_test_data_loader_list = \
DataPreProcessor.preprocess_texture_except_DTD(texture_train_data_set_path,
texture_train_label_set_path,
train_parameters[
"texture_batch_size"],
num_workers=0,
device=device,
split_size=split_size,
type="UIUC")
image_net_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_data_set_path,
image_net_label_set_path,
train_parameters[
"image_net_batch_size"],
image_net_test_path,
num_workers=0,
split_size=split_size,
device=device,
type="ImageNet")
train_arguments = {
"IMAGE_NET_LABELS": IMAGE_NET_LABELS,
"TEXTURE_LABELS": TEXTURE_LABELS,
"image_net_data_loader_dict": image_net_data_loader_dict,
"texture_data_loader_list": texture_train_val_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
network = train.train_classifier(train_arguments, device, dataset_name="UIUC")
# training ends
# testing starts
model_path_bn = "./Models/MTL/UIUC/Multitask_Classifier_Model_epoch_75_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": texture_test_data_loader_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
test = Test_Classifier()
accuracy_list, mean_accuracy = test.test_classifier(test_arguments, IMAGE_NET_LABELS, device,
dataset_name="UIUC")
file1 = open("UIUC_Details.txt", "a")
file1.write(str(train_parameters))
file1.write("UIUC Mean accuracy: {0}\n".format(mean_accuracy))
file1.write(str(accuracy_list))
pd.DataFrame.from_dict(
accuracy_list,
orient='columns'
).to_csv("./Accuracy_UIUC.csv")
def kth_Train_test_single():
# 0.05 test train val split
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ["KTH_aluminium_foil", "KTH_brown_bread", "KTH_corduroy", "KTH_cork",
"KTH_cotton",
"KTH_cracker", "KTH_linen",
"KTH_orange_peel", "KTH_sponge", "KTH_styrofoam", "KTH_wool"]
train_parameters = {
"epochs": 25,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
texture_train_data_set_path = "./Dataset/Texture/kth/kth_X.pickle"
texture_train_label_set_path = "./Dataset/Texture/kth/kth_Y.pickle"
saved_model_name = "./Models/MTL/kth/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.20
# training starts
texture_train_val_data_loader_list, texture_test_data_loader_list = \
DataPreProcessor.preprocess_texture_except_DTD(texture_train_data_set_path,
texture_train_label_set_path,
train_parameters[
"texture_batch_size"],
num_workers=0,
device=device,
split_size=split_size,
type="Kth",
folds=1)
train_arguments = {
"TEXTURE_LABELS": TEXTURE_LABELS,
"texture_data_loader_list": texture_train_val_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
train.train_classifier_single(train_arguments, device)
# training ends
# testing starts
model_path_bn = "./Models/MTL/kth/Multitask_Classifier_Model_epoch_25_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": texture_test_data_loader_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
test = Test_Classifier()
accuracy_list, mean_accuracy = test.test_classifier_single(test_arguments, device)
file1 = open("kth_Details.txt", "a")
file1.write(str(train_parameters))
file1.write("kth Mean accuracy: {0}\n".format(mean_accuracy))
file1.write(str(accuracy_list))
pd.DataFrame.from_dict(
accuracy_list,
orient='columns'
).to_csv("./Accuracy_Kth_single.csv")
def surface_Train_test():
# 0.05 test train val split
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ['Kyberge_blanket1', 'Kyberge_canvas1', 'Kyberge_seat2', 'UIUC07_water',
'UIUC02_bark2', 'KTH_brown_bread', 'UIUC17_glass2',
'Kyberge_scarf1', 'KTH_corduroy', 'UIUC16_glass1', 'Kyberge_stoneslab1',
'Kyberge_rice2', 'UIUC06_wood3', 'KTH_aluminium_foil', 'Kyberge_ceiling1',
'Kyberge_sesameseeds1', 'Kyberge_floor2', 'Kyberge_lentils1', 'KTH_linen',
'UIUC08_granite', 'Kyberge_screen1', 'UIUC24_corduroy', 'Kyberge_oatmeal1',
'Kyberge_stone1', 'UIUC03_bark3', 'Kyberge_pearlsugar1', 'UIUC05_wood2',
'UIUC14_brick1', 'UIUC19_carpet2', 'UIUC23_knit', 'UIUC22_fur', 'UIUC15_brick2',
'KTH_wool', 'KTH_orange_peel', 'Kyberge_blanket2', 'Kyberge_sand1', 'KTH_sponge',
'Kyberge_seat1', 'Kyberge_scarf2', 'KTH_cracker', 'Kyberge_grass1', 'Kyberge_rice1',
'KTH_cork', 'UIUC04_wood1', 'Kyberge_cushion1', 'Kyberge_stone3', 'UIUC18_carpet1',
'Kyberge_ceiling2', 'UIUC10_floor1', 'Kyberge_floor1', 'Kyberge_stone2', 'KTH_cotton',
'UIUC09_marble', 'Kyberge_wall1', 'Kyberge_linseeds1', 'UIUC12_pebbles', 'UIUC11_floor2',
'UIUC01_bark1', 'Kyberge_rug1', 'KTH_styrofoam', 'UIUC25_plaid', 'UIUC21_wallpaper',
'UIUC13_wall',
'UIUC20_upholstery']
IMAGE_NET_LABELS = \
["alp", "artichoke", "Band Aid", "bathing cap", "bookshop", "bull mastiff", "butcher shop",
"carbonara", "chain", "chain saw", "chainlink fence", "cheetah", "cliff dwelling", "common iguana",
"confectionery", "container ship", "corn", "crossword puzzle", "dishrag", "dock", "flat-coated retriever",
"gibbon", "grocery store", "head cabbage", "honeycomb", "hook", "hourglass", "jigsaw puzzle",
"jinrikisha", "lakeside", "lawn mower", "maillot", "microwave", "miniature poodle", "muzzle",
"notebook", "ocarina", "orangutan", "organ", "paper towel", "partridge", "rapeseed",
"sandbar", "sarong", "sea urchin", "shoe shop", "shower curtain", "stone wall", "theater curtain",
"tile roof",
"turnstile", "vault", "velvet", "window screen", "wool", "yellow lady's slipper"]
train_parameters = {
"epochs": 1000,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
image_net_data_set_path = "./Dataset/ImageNet/ImageNet_X.pickle"
image_net_label_set_path = "./Dataset/ImageNet/ImageNet_Y.pickle"
image_net_test_path = "./Dataset/ImageNet/ImageNet_Test.pickle"
texture_train_data_set_path = "./Dataset/Texture/Surface/Surface_X_train.pickle"
texture_train_label_set_path = "./Dataset/Texture/Surface/Surface_Y_train.pickle"
texture_test_data_set_path = "./Dataset/Texture/Surface/Surface_X_vaild.pickle"
texture_test_label_set_path = "./Dataset/Texture/Surface/Surface_Y_vaild.pickle"
saved_model_name = "./Models/MTL/Surface/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.20
# training starts
texture_train_val_data_loader_list, texture_test_data_loader_list = \
DataPreProcessor.preprocess_texture_surface(texture_train_data_set_path,
texture_train_label_set_path,
texture_test_data_set_path,
texture_test_label_set_path,
train_parameters["texture_batch_size"],
num_workers=0,
device=device, split_size=split_size,
type="Surface", folds=1)
image_net_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_data_set_path,
image_net_label_set_path,
train_parameters[
"image_net_batch_size"],
image_net_test_path,
num_workers=0,
split_size=split_size,
device=device,
type="ImageNet")
train_arguments = {
"IMAGE_NET_LABELS": IMAGE_NET_LABELS,
"TEXTURE_LABELS": TEXTURE_LABELS,
"image_net_data_loader_dict": image_net_data_loader_dict,
"texture_data_loader_list": texture_train_val_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
network = train.train_classifier(train_arguments, device, dataset_name="kth")
# training ends
# testing starts
model_path_bn = "./Models/MTL/Surface/Multitask_Classifier_Model_epoch_400_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": texture_test_data_loader_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
test = Test_Classifier()
accuracy_list, mean_accuracy = test.test_classifier(test_arguments, IMAGE_NET_LABELS, device)
file1 = open("Surface_Details.txt", "a")
file1.write(str(train_parameters))
file1.write("Surface Mean accuracy: {0}\n".format(mean_accuracy))
file1.write(str(accuracy_list))
pd.DataFrame.from_dict(
accuracy_list,
orient='columns'
).to_csv("./Accuracy_Surface.csv")
def DTD_Train_test():
device = Util.get_device()
# device = 'cuda' if torch.cuda.is_available() else 'cpu'
print(device)
TEXTURE_LABELS = ["banded", "blotchy", "braided", "bubbly", "bumpy", "chequered", "cobwebbed", "cracked",
"crosshatched", "crystalline",
"dotted", "fibrous", "flecked", "freckled", "frilly", "gauzy", "grid", "grooved",
"honeycombed",
"interlaced", "knitted",
"lacelike", "lined", "marbled", "matted", "meshed", "paisley", "perforated", "pitted",
"pleated",
"polka-dotted", "porous",
"potholed", "scaly", "smeared", "spiralled", "sprinkled", "stained", "stratified", "striped",
"studded", "swirly", "veined",
"waffled", "woven", "wrinkled", "zigzagged"]
IMAGE_NET_LABELS = \
["alp", "artichoke", "Band Aid", "bathing cap", "bookshop", "bull mastiff", "butcher shop",
"carbonara", "chain", "chain saw", "chainlink fence", "cheetah", "cliff dwelling", "common iguana",
"confectionery", "container ship", "corn", "crossword puzzle", "dishrag", "dock", "flat-coated retriever",
"gibbon", "grocery store", "head cabbage", "honeycomb", "hook", "hourglass", "jigsaw puzzle",
"jinrikisha", "lakeside", "lawn mower", "maillot", "microwave", "miniature poodle", "muzzle",
"notebook", "ocarina", "orangutan", "organ", "paper towel", "partridge", "rapeseed",
"sandbar", "sarong", "sea urchin", "shoe shop", "shower curtain", "stone wall", "theater curtain",
"tile roof",
"turnstile", "vault", "velvet", "window screen", "wool", "yellow lady's slipper"]
IMAGE_NET_LABELS_S2 = \
["common iguana", "partridge", "flat-coated retriever", "bull mastiff", "miniature poodle", "cheetah",
"sea urchin", "orangutan", "gibbon", "Band Aid", "bathing cap", "chain saw", "container ship", "hook",
"hourglass", "jinrikisha", "lawn mower", "maillot", "microwave", "muzzle", "notebook", "ocarina", "organ",
"paper towel", "sarong", "turnstile", "crossword puzzle", "yellow lady's slipper"
]
IMAGE_NET_LABELS_T = \
["alp", "artichoke", "bookshop", "butcher shop",
"carbonara", "chain", "chainlink fence", "cliff dwelling",
"confectionery", "corn", "dishrag", "dock",
"grocery store", "head cabbage", "honeycomb", "jigsaw puzzle",
"lakeside", "rapeseed",
"sandbar", "shoe shop", "shower curtain", "stone wall", "theater curtain", "tile roof",
"vault", "velvet", "window screen", "wool", ]
train_parameters = {
"epochs": 400,
"learning_rate": 0.0001,
"texture_batch_size": 32,
"image_net_batch_size": 32,
"weight_decay": 0.0005
}
image_net_data_set_path = "./Dataset/ImageNet/ImageNet_X.pickle"
image_net_label_set_path = "./Dataset/ImageNet/ImageNet_Y.pickle"
image_net_test_path = "./Dataset/ImageNet/ImageNet_Test.pickle"
image_net_S2_data_set_path = "./Dataset/ImageNet/ImageNet_S2X.pickle"
image_net_S2_label_set_path = "./Dataset/ImageNet/ImageNet_S2Y.pickle"
image_net_S2_test_path = "./Dataset/ImageNet/ImageNet_S2_Test.pickle"
image_net_T_data_set_path = "./Dataset/ImageNet/ImageNet_TX.pickle"
image_net_T_label_set_path = "./Dataset/ImageNet/ImageNet_TY.pickle"
image_net_T_test_path = "./Dataset/ImageNet/ImageNet_T_Test.pickle"
texture_train_data_set_path = "./Dataset/Texture/DTD/Texture_DTD_train{0}_X.pickle"
texture_train_label_set_path = "./Dataset/Texture/DTD/Texture_DTD_train{0}_Y.pickle"
texture_val_data_set_path = "./Dataset/Texture/DTD/Texture_DTD_val{0}_X.pickle"
texture_val_label_set_path = "./Dataset/Texture/DTD/Texture_DTD_val{0}_Y.pickle"
auto_encoder_model_path = "./Models/Auto_encoder_Model_epoch_300_lr_0.001_noise_factor_0.5.pt"
saved_model_name = "./Models/MTL/DTD/Multitask_Classifier_Model_epoch_" + str(
train_parameters["epochs"]) + "_lr_" + str(
train_parameters["learning_rate"]) + "_split{0}.pth"
split_size = 0.05
# training starts
texture_data_loader_list = DataPreProcessor.preprocess_DTD_train_val_10_splits(texture_train_data_set_path,
texture_train_label_set_path,
texture_val_data_set_path,
texture_val_label_set_path,
train_parameters[
"texture_batch_size"],
num_workers=0,
device=device)
# image_net_S2_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_S2_data_set_path,
# image_net_S2_label_set_path,
# train_parameters[
# "image_net_batch_size"],
# image_net_S2_test_path,
# num_workers=0,
# split_size=split_size,
# device=device,
# type="ImageNet_S2"
# )
# image_net_T_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_T_data_set_path,
# image_net_T_label_set_path,
# train_parameters[
# "image_net_batch_size"],
# image_net_T_test_path,
# num_workers=0,
# split_size=split_size,
# device=device,
# type="ImageNet_T"
# )
image_net_data_loader_dict = DataPreProcessor.preprocess_image_net(image_net_data_set_path,
image_net_label_set_path,
train_parameters[
"image_net_batch_size"],
image_net_test_path,
num_workers=0,
split_size=split_size,
device=device,
type="ImageNet")
train_arguments = {
"IMAGE_NET_LABELS": IMAGE_NET_LABELS,
"IMAGE_NET_LABELS_S2": IMAGE_NET_LABELS_S2,
"IMAGE_NET_LABELS_T": IMAGE_NET_LABELS_T,
"TEXTURE_LABELS": TEXTURE_LABELS,
# "image_net_S2_data_loader_dict": image_net_S2_data_loader_dict,
# "image_net_T_data_loader_dict": image_net_T_data_loader_dict,
"image_net_data_loader_dict": image_net_data_loader_dict,
"texture_data_loader_list": texture_data_loader_list,
"train_parameters": train_parameters,
"saved_model_name": saved_model_name
}
train = Train_Classifier()
network = train.train_classifier(train_arguments, device, dataset_name="DTD")
# training ends
# testing starts
texture_data_set_path = "./Dataset/Texture/DTD/Texture_DTD_test{0}_X.pickle"
texture_label_set_path = "./Dataset/Texture/DTD/Texture_DTD_test{0}_Y.pickle"
data_loader_test_list = DataPreProcessor.prepare_data_loader_test_10_splits(texture_data_set_path,
texture_label_set_path,
device)
model_path_bn = "./Models/MTL/DTD/Multitask_Classifier_Model_epoch_400_lr_0.0001_split{0}.pth"
test_arguments = {
"data_loader_test_list": data_loader_test_list,
"model_path_bn": model_path_bn,
"TEXTURE_LABELS": TEXTURE_LABELS
}
# test(test_arguments, IMAGE_NET_LABELS, device)
test = Test_Classifier()
test.test_classifier(test_arguments, IMAGE_NET_LABELS, device, dataset_name="DTD")