BenchmarkTransferLearning_f/main_classification.py at main · felixkrones/BenchmarkTransferLearning_f · GitHub

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
import os
import sys
import shutil
import time
import numpy as np
from optparse import OptionParser
from shutil import copyfile
from tqdm import tqdm

from utils import vararg_callback_bool, vararg_callback_int
from dataloader import  *

import torch
from engine import classification_engine

sys.setrecursionlimit(40000)


def get_args_parser(main_args:bool=True):
    parser = OptionParser()

    parser.add_option("--GPU", dest="GPU", help="the index of gpu is used", default=None, action="callback",
                      callback=vararg_callback_int)
    parser.add_option("--model", dest="model_name", help="DenseNet121| vit_base| swin_base", default="Resnet50", type="string")
    parser.add_option("--init", dest="init",
                      help="Random| ImageNet| ImageNet_1k| ImageNet_21k| SAM| DeiT| BEiT| DINO| MoCo_V3| GMML| MoBY | MAE| SimMIM",
                      default='', type="string")
    parser.add_option("--num_class", dest="num_class", help="number of the classes in the downstream task",
                      default=14, type="int")
    parser.add_option("--data_set", dest="data_set", help="ChestXray14|CheXpert|padchest", default="ChestXray14", type="string")
    parser.add_option("--normalization", dest="normalization", help="how to normalize data (imagenet|chestx-ray)", default="imagenet",
                      type="string")
    parser.add_option("--img_size", dest="img_size", help="input image resolution", default=224, type="int")
    parser.add_option("--nc", dest="nc", help="num of image channels", default=1, type="int")
    parser.add_option("--img_depth", dest="img_depth", help="num of image depth", default=3, type="int")
    parser.add_option("--data_dir", dest="data_dir", help="dataset dir",default="data/", type="string")
    parser.add_option("--train_list", dest="train_list", help="file for training list",
                      default=None, type="string")
    parser.add_option("--val_list", dest="val_list", help="file for validating list",
                      default=None, type="string")
    parser.add_option("--test_list", dest="test_list", help="file for test list",
                      default=None, type="string")
    parser.add_option("--mode", dest="mode", help="train | test", default="train", type="string")
    parser.add_option("--batch_size", dest="batch_size", help="batch size", default=64, type="int")
    parser.add_option("--epochs", dest="num_epoch", help="num of epoches", default=30, type="int")

    # Optimizer parameters
    parser.add_option("--optimizer", dest="optimizer", help="Adam | SGD", default="Adam", type="string")
    parser.add_option('--opt', default='adamw', type=str, metavar='OPTIMIZER',
                        help='Optimizer (default: "adamw"')
    parser.add_option('--opt-eps', default=1e-8, type=float, metavar='EPSILON',
                        help='Optimizer Epsilon (default: 1e-8)')
    parser.add_option('--opt-betas', default=None, type=float, nargs='+', metavar='BETA',
                        help='Optimizer Betas (default: None, use opt default)')
    parser.add_option('--clip-grad', type=float, default=None, metavar='NORM',
                        help='Clip gradient norm (default: None, no clipping)')
    parser.add_option('--momentum', type=float, default=0.9, metavar='M',
                        help='SGD momentum (default: 0.9)')
    parser.add_option('--weight-decay', type=float, default=0.05,
                    help='weight decay (default: 0.05)')
    parser.add_option('--momentum-decay', type=float, default=0.0,
                        help='weight decay (default: 0.05)')

    # Learning rate schedule parameters
    parser.add_option('--sched', default='cosine', type=str, metavar='SCHEDULER',
                        help='LR scheduler (default: "cosine"')
    parser.add_option("--lr", dest="lr", help="learning rate", default=2.5e-6, type="float")
    parser.add_option("--lr_Scheduler", dest="lr_Scheduler", help="learning schedule", default="ReduceLROnPlateau",
                      type="string")
    parser.add_option('--lr-noise', type=float, nargs='+', default=None, metavar='pct, pct',
                        help='learning rate noise on/off epoch percentages')
    parser.add_option('--lr-noise-pct', type=float, default=0.67, metavar='PERCENT',
                        help='learning rate noise limit percent (default: 0.67)')
    parser.add_option('--lr-noise-std', type=float, default=1.0, metavar='STDDEV',
                        help='learning rate noise std-dev (default: 1.0)')
    parser.add_option('--warmup-lr', type=float, default=1e-6, metavar='LR',
                        help='warmup learning rate (default: 1e-6)')
    parser.add_option('--min-lr', type=float, default=5e-7, metavar='LR',
                        help='lower lr bound for cyclic schedulers that hit 0 (1e-5)')
    parser.add_option('--decay-epochs', type=float, default=30, metavar='N',
                        help='epoch interval to decay LR')
    parser.add_option('--warmup-epochs', type=int, default=5, metavar='N',
                        help='epochs to warmup LR, if scheduler supports')
    parser.add_option('--cooldown-epochs', type=int, default=10, metavar='N',
                        help='epochs to cooldown LR at min_lr, after cyclic schedule ends')
    parser.add_option('--patience-epochs', type=int, default=10, metavar='N',
                         help='patience epochs for Plateau LR scheduler (default: 10')
    parser.add_option('--decay-rate', '--dr', type=float, default=0.5, metavar='RATE',
                        help='LR decay rate (default: 0.1)')

    parser.add_option("--patience", dest="patience", help="num of patient epoches", default=10, type="int")
    parser.add_option("--early_stop", dest="early_stop", help="whether use early_stop", default=True, action="callback",
                      callback=vararg_callback_bool)
    parser.add_option("--trial", dest="num_trial", help="number of trials", default=5, type="int")
    parser.add_option("--start_index", dest="start_index", help="the start model index", default=0, type="int")
    parser.add_option("--clean", dest="clean", help="clean the existing data", default=False, action="callback",
                      callback=vararg_callback_bool)
    parser.add_option("--resume", dest="resume", help="whether latest checkpoint", default=True, action="callback",
                      callback=vararg_callback_bool)
    parser.add_option("--best", dest="best", help="whether to use last or best checkpoint", default="last", action="callback",
                      callback=vararg_callback_bool)
    parser.add_option("--workers", dest="workers", help="number of CPU workers", default=8, type="int")
    parser.add_option("--print_freq", dest="print_freq", help="print frequency", default=50, type="int")
    parser.add_option("--test_augment", dest="test_augment", help="whether use test time augmentation",
                      default=False, action="callback", callback=vararg_callback_bool)
    parser.add_option("--proxy_dir", dest="proxy_dir", help="Path to the Pretrained model", default=None, type="string")
    parser.add_option("--label_layers", default=0, type=int, help='For GMML, how many label token layers. 12 means introduce LT at the first layer')
    parser.add_option("--anno_percent", dest="anno_percent", help="data percent", default=100, type="int")
    parser.add_option("--device", dest="device", help="cpu|cuda", default="cuda", type="string")
    parser.add_option("--activate", dest="activate", help="Sigmoid", default="Sigmoid", type="string")
    parser.add_option("--uncertain_label", dest="uncertain_label",
                      help="the label assigned to uncertain data (Ones | Zeros | LSR-Ones | LSR-Zeros)",
                      default="Zeros", type="string")
    parser.add_option("--unknown_label", dest="unknown_label", help="the label assigned to unknown data",
                      default=0, type="int")

    if main_args:
        (options, args) = parser.parse_args()
        return options
    else:
        return parser


def main(args):
    print(args)
    assert args.data_dir is not None
    assert args.train_list is not None
    assert args.val_list is not None
    assert args.test_list is not None
    if args.init.lower() != 'imagenet' and args.init.lower() != 'random':
        assert args.proxy_dir is not None

    if args.proxy_dir is not None:
        args.exp_name = args.proxy_dir.split("/")[-1].split(".")[0]
    else:
        args.exp_name = args.model_name + "_" + args.init
    model_path = os.path.join("./models/Classification",args.data_set)
    output_path = os.path.join("./Outputs/Classification",args.data_set)
    test_diseases_name = ["Atelectasis", "Cardiomegaly", "Consolidation", "Edema", "Pneumonia", "Pneumothorax"]

    if args.data_set == "ChestXray14":
        diseases = ['Atelectasis', 'Cardiomegaly', 'Effusion', 'Infiltration', 'Mass', 'Nodule',
                    'Pneumonia', 'Pneumothorax', 'Consolidation', 'Edema',
                    'Emphysema', 'Fibrosis', 'Pleural_Thickening', 'Hernia']
        dataset_train = ChestXray14Dataset(images_path=args.data_dir, file_path=args.train_list,augment=build_transform_classification(normalize=args.normalization, mode="train", nc=args.nc), nc=args.nc)
        dataset_val = ChestXray14Dataset(images_path=args.data_dir, file_path=args.val_list,augment=build_transform_classification(normalize=args.normalization, mode="valid", nc=args.nc), nc=args.nc)
        dataset_test = ChestXray14Dataset(images_path=args.data_dir, file_path=args.test_list,augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment, nc=args.nc), nc=args.nc)

        classification_engine(args, model_path, output_path, diseases, dataset_train, dataset_val, dataset_test)


    elif args.data_set == "CheXpert":
        diseases = ['No Finding', 'Enlarged Cardiomediastinum', 'Cardiomegaly', 'Lung Opacity',
                           'Lung Lesion', 'Edema', 'Consolidation', 'Pneumonia', 'Atelectasis', 'Pneumothorax',
                           'Pleural Effusion', 'Pleural Other', 'Fracture', 'Support Devices']
        #test_diseases_name = ['Atelectasis', 'Cardiomegaly', 'Consolidation', 'Edema', 'Pleural Effusion']
        test_diseases = [diseases.index(c) for c in test_diseases_name]
        dataset_train = CheXpertDataset(images_path=args.data_dir, file_path=args.train_list,
                                        augment=build_transform_classification(normalize=args.normalization, mode="train", nc=args.nc), uncertain_label=args.uncertain_label, unknown_label=args.unknown_label, annotation_percent=args.anno_percent, num_class=len(diseases), nc=args.nc)
        dataset_val = CheXpertDataset(images_path=args.data_dir, file_path=args.val_list,
                                      augment=build_transform_classification(normalize=args.normalization, mode="valid", nc=args.nc), uncertain_label=args.uncertain_label, unknown_label=args.unknown_label, nc=args.nc)
        dataset_test = CheXpertDataset(images_path=args.data_dir, file_path=args.test_list,
                                       augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment, nc=args.nc), uncertain_label=args.uncertain_label, unknown_label=args.unknown_label, nc=args.nc)

        print(f"Got dataset {args.data_set}, starting classification_engine.")
        classification_engine(args, model_path, output_path, diseases, dataset_train, dataset_val, dataset_test, test_diseases)


    elif args.data_set == "Shenzhen":
        diseases = ['TB']
        dataset_train = ShenzhenCXR(images_path=args.data_dir, file_path=args.train_list,
                                    augment=build_transform_classification(normalize=args.normalization, mode="train"), annotation_percent=args.anno_percent)

        dataset_val = ShenzhenCXR(images_path=args.data_dir, file_path=args.val_list,
                                  augment=build_transform_classification(normalize=args.normalization, mode="valid"), annotation_percent=args.anno_percent)

        dataset_test = ShenzhenCXR(images_path=args.data_dir, file_path=args.test_list,
                                   augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment), annotation_percent=args.anno_percent)


        classification_engine(args, model_path, output_path, diseases, dataset_train, dataset_val, dataset_test)


    elif args.data_set == "padchest":
        dataset_train = PadchestDataset(images_path=args.data_dir, file_path=args.train_list, augment=build_transform_classification(normalize=args.normalization, mode="train"), possible_labels=test_diseases_name)
        dataset_val = PadchestDataset(images_path=args.data_dir, file_path=args.val_list, augment=build_transform_classification(normalize=args.normalization, mode="valid"), possible_labels=test_diseases_name)
        dataset_test = PadchestDataset(images_path=args.data_dir, file_path=args.test_list, augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment), possible_labels=test_diseases_name)

        classification_engine(args, model_path, output_path, test_diseases_name, dataset_train, dataset_val, dataset_test)

    elif args.data_set == "MIMIC":
        dataset_train = MIMIC_Dataset(images_path=args.data_dir, file_path=args.train_list, augment=build_transform_classification(normalize=args.normalization, mode="train"), n_samples=100000)
        dataset_val = MIMIC_Dataset(images_path=args.data_dir, file_path=args.val_list, augment=build_transform_classification(normalize=args.normalization, mode="valid"), n_samples=5000)
        dataset_test = MIMIC_Dataset(images_path=args.data_dir, file_path=args.test_list, augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment), n_samples=5000)

        classification_engine(args, model_path, output_path, None, dataset_train, dataset_val, dataset_test)

    elif args.data_set == "COCO":
        dataset_train = COCO(images_path=args.data_dir, file_path=args.train_list, augment=build_transform_classification(normalize=args.normalization, mode="train"), n_samples=100000)
        dataset_val = COCO(images_path=args.data_dir, file_path=args.val_list, augment=build_transform_classification(normalize=args.normalization, mode="valid"), n_samples=5000)
        dataset_test = COCO(images_path=args.data_dir, file_path=args.test_list, augment=build_transform_classification(normalize=args.normalization, mode="test", test_augment=args.test_augment), n_samples=5000)

        classification_engine(args, model_path, output_path, None, dataset_train, dataset_val, dataset_test)

    else:
        raise NotImplementedError(f"Dataset {args.data_set} not implemented.")


if __name__ == '__main__':
    args = get_args_parser()
    args.epochs = args.num_epoch
    main(args)