InsightFace_Pytorch/train.py at master · luzai/InsightFace_Pytorch · 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
#!/usr/bin/env python3

print('------------------------')
import os, time

try:
    import install
    from lz import *

    try:
        import moxing, moxing.pytorch as mox

        moxing.file.shift('os', 'mox')
    except:
        pass
    import torch
    from config import conf
    from pathlib import Path
    from exargs import parser

    torch.backends.cudnn.benchmark = True
    torch.backends.cudnn.deterministic = False
    torch.backends.cudnn.enabled = True

    try:
        import mox_patch_0603_v4
        import moxing.pytorch as mox
    except:
        logging.warning('not in the cloud')
        conf.cloud = False
except Exception as f:
    print(f)
    time.sleep(600)


def main():
    args = parser.parse_args()
    print('args.data_url', args.data_url)
    if conf.cloud:
        mox.file.copy_parallel(args.data_url, '/cache/face_train/')
        args.data_url = '/cache/face_train/'
        conf.use_data_folder = args.data_url

    if args.work_path:
        conf.work_path = Path(args.work_path)
        conf.model_path = conf.work_path / 'models'
        conf.log_path = conf.work_path / 'log'
        conf.save_path = conf.work_path / 'save'
    else:
        args.work_path = conf.work_path
    conf.update(args.__dict__)
    if conf.local_rank is not None:
        torch.cuda.set_device(conf.local_rank)
        torch.distributed.init_process_group(backend='nccl',
                                             init_method="env://")
        if torch.distributed.get_rank() != 0:
            set_stream_logger(logging.WARNING)
    # if osp.exists(conf.save_path):
    #     logging.info('ok')
    #     exit(1)
    # simplify_conf(conf)
    # exit(0)
    from Learner import face_learner
    # decs = msgpack_load('decs.pk')
    # conf.decs = decs
    learner = face_learner(conf, )
    # fstrs = learner.list_fixed_strs('work_space/sglpth.casia/models')
    # stps = learner.list_steps('work_space/sglpth.casia/models')
    # fstr = fstrs[np.argmax(stps)]
    # stt_dct = torch.load('work_space/sglpth.casia/models/model_' + fstr)
    # learner.model.module.load_state_dict_sglpth(stt_dct)
    # print(fstrs, stps, fstr, )

    if conf.get('load_from'):
        # p= 'r100.128.retina.clean.arc',
        # 'hrnet.retina.arc.3',
        # 'mbv3.retina.arc',
        # 'mbfc.lrg.retina.arc.s48',
        # 'effnet.casia.arc',
        # 'mbfc.retina.cl.distill.cont2',
        # 'mbfc2',
        # 'r18.l2sft',
        # 'r18.adamrg',
        # 'mbfc.se.elu.ms1m.radam.1',
        # 'mbfc.se.elu.specnrm.allbutdw.ms1m.adam.1',
        # 'mbfc.se.prelu.specnrm.ms1m.cesigsft.1',
        # 'irse.elu.ms1m',
        # 'irse.elu.casia.arc.2048',
        p = Path(conf.load_from)
        print('try to load from ', p, )
        learner.load_state(
            resume_path=p,
            load_optimizer=False,
            load_head=conf.head_load,  # todo note!
            load_imp=False,
            latest=True, strict=False,
        )
    # simplify_conf(conf)
    learner.cloud_sync_log()
    # res = learner.validate_ori(conf, valds_names=('cfp_fp', ))
    # exit(0)
    # learner.calc_img_feas(out='work_space/mbfc.crash.h5')
    # log_lrs, losses = learner.find_lr(
    #                                   num=999,
    #                                   bloding_scale=1000)
    # losses[np.isnan(losses)] = 999
    # best_lr = 10 ** (log_lrs[np.argmin(losses)])
    # print('best lr is ', best_lr)
    # conf.lr = best_lr
    # exit(0)

    # learner.init_lr()
    # conf.tri_wei = 0
    # log_conf(conf)
    # learner.train(conf, 1, name='xent')

    learner.init_lr()
    simplify_conf(conf)
    if conf.head_init:
        learner.head_initialize()
    if conf.warmup:
        learner.warmup(conf, conf.warmup)
    learner.train_simple(conf, conf.epochs)

    # learner.train_dist(conf, conf.epochs)
    if conf.net_mode == 'sglpth':
        decs = learner.model.module.get_decisions()
        msgpack_dump(decs, 'decs.pk')

    # learner.train_cotching(conf, conf.epochs)
    # learner.train_cotching_accbs(conf, conf.epochs)
    # learner.train_ghm(conf, conf.epochs)
    # learner.train_with_wei(conf, conf.epochs)
    # learner.train_use_test(conf, conf.epochs)

    # res = learner.validate_ori(conf, )
    if not conf.cloud:
        from tools.test_ijbc3 import test_ijbc3
        res = test_ijbc3(conf, learner)
        tpr6, tpr4, tpr3 = res[0][1], res[1][1], res[2][1]
        learner.writer.add_scalar('ijbb/6', tpr6, learner.step)
        learner.writer.add_scalar('ijbb/4', tpr4, learner.step)
        learner.writer.add_scalar('ijbb/3', tpr3, learner.step)
    learner.writer.close()

    if conf.never_stop:
        img = torch.randn((conf.batch_size // 2, 3, conf.input_size, conf.input_size)).cuda()
        learner.model.eval()
        logging.info('never stop')
        while True:
            _ = learner.model(img)


if __name__ == '__main__':
    # try:
    main()
# except Exception as f:
#     print(f)
#     time.sleep(600)