forked from yzhou359/MakeItTalk
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main_train_content.py
136 lines (104 loc) · 5.79 KB
/
main_train_content.py
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
"""
# Copyright 2020 Adobe
# All Rights Reserved.
# NOTICE: Adobe permits you to use, modify, and distribute this file in
# accordance with the terms of the Adobe license agreement accompanying
# it.
"""
import os, glob
import numpy as np
import cv2
import argparse
import platform
import torch
from util.utils import try_mkdir
from src.approaches.train_content import Audio2landmark_model
if platform.release() == '4.4.0-83-generic':
ROOT_DIR = r'/mnt/ntfs/Dataset/TalkingToon/Obama_for_train'
else: # 3.10.0-957.21.2.el7.x86_64
ROOT_DIR = r'/mnt/nfs/work1/kalo/yangzhou/TalkingToon/VoxCeleb2'
DEMO_CH = ''
parser = argparse.ArgumentParser()
parser.add_argument('--nepoch', type=int, default=1001, help='number of epochs to train for')
parser.add_argument('--batch_size', type=int, default=1, help='batch size')
parser.add_argument('--in_batch_nepoch', type=int, default=1, help='')
parser.add_argument('--first_in_batch_nepoch', type=int, default=1, help='')
parser.add_argument('--segment_batch_size', type=int, default=128, help='batch size')
parser.add_argument('--num_window_frames', type=int, default=18, help='')
parser.add_argument('--num_window_frames_seq', type=int, default=18, help='')
parser.add_argument('--num_window_frames_sync', type=int, default=18, help='')
parser.add_argument('--num_window_step', type=int, default=1, help='')
parser.add_argument('--dump_dir', type=str, default='', help='')
parser.add_argument('--dump_file_name', type=str, default='celeb_withrot', help='')
parser.add_argument('--lr', type=float, default=1e-3, help='learning rate')
parser.add_argument('--reg_lr', type=float, default=0., help='weight decay')
parser.add_argument('--drop_out', type=float, default=0, help='drop out')
parser.add_argument('--verbose', type=int, default=1, help='0 - detail, 2 - simplify')
parser.add_argument('--write', default=False, action='store_true')
parser.add_argument('--add_pos', default=False, action='store_true')
parser.add_argument('--use_motion_loss', default=False, action='store_true')
parser.add_argument('--name', type=str, default='tmp')
parser.add_argument('--puppet_name', type=str, default=DEMO_CH)
parser.add_argument('--in_size', type=int, default=80)
parser.add_argument('--use_lip_weight', default=False, action='store_true')
parser.add_argument('--use_adain', default=False, action='store_true')
parser.add_argument('--use_residual', default=True, action='store_true')
parser.add_argument('--use_norm_emb', default=False, action='store_true')
parser.add_argument('--use_cycle_loss', default=False, action='store_true')
parser.add_argument('--lambda_cycle_loss', default=1.0, type=float)
parser.add_argument('--emb_coef', default=3.0, type=float)
parser.add_argument('--freeze_content_emb', default=False, action='store_true')
parser.add_argument('--pretrain_g', default=False, action='store_true')
parser.add_argument('--spk_emb_enc_size', default=16, type=int)
parser.add_argument('--c_enc_hidden_size', default=256, type=int)
parser.add_argument('--lstm_g_hidden_size', default=256, type=int)
parser.add_argument('--projection_size', default=512, type=int)
parser.add_argument('--use_addinfo_format', default='motion_and_pos')
parser.add_argument('--l2_on_fls_without_traj', default=False, action='store_true')
parser.add_argument('--train_with_grad_penalty', default=False, action='store_true')
parser.add_argument('--train_DL', default=-1.0, type=float)
parser.add_argument('--train_DT', default=-1.0, type=float)
parser.add_argument('--train_G_only', default=False, action='store_true')
parser.add_argument('--lambda_mse_loss', default=1.0, type=float)
parser.add_argument('--teacher_force', default=0.0, type=float)
parser.add_argument('--debug_version', default='', type=str)
parser.add_argument('--lambda_add_info_loss', default=1.0, type=float)
parser.add_argument('--show_animation', default=False, action='store_true')
# model
parser.add_argument('--pos_dim', default=7, type=int)
parser.add_argument('--use_prior_net', default=True, action='store_true')
parser.add_argument('--transformer_d_model', default=32, type=int)
parser.add_argument('--transformer_N', default=2, type=int)
parser.add_argument('--transformer_heads', default=2, type=int)
parser.add_argument('--load_a2l_C_name', type=str,
default='/mnt/ntfs/Dataset/TalkingToon/Obama_for_train/ckpt/local_obama_norm_4_batch2/ckpt_e_100.pth') # 'examples/ckpt/ckpt_audio2landmark_c.pth')
# data
parser.add_argument('--use_11spk_only', default=True, action='store_true')
# arch
parser.add_argument('--use_reg_as_std', default=True, action='store_false')
parser.add_argument('--lambda_laplacian_smooth_loss', default=1.0, type=float)
# test
parser.add_argument('--test_emb', default=False, action='store_true')
parser.add_argument('--train', default=False, action='store_true')
parser.add_argument('--test_end2end', default=False, action='store_true')
# save model
parser.add_argument('--jpg_freq', type=int, default=10, help='')
parser.add_argument('--ckpt_epoch_freq', type=int, default=10, help='')
AMP = {'default':[2.5, 2.5, 1.0]}
if(DEMO_CH not in AMP.keys()):
AMP[DEMO_CH] = AMP['default']
parser.add_argument('--amp_lip_x', type=float, default=AMP[DEMO_CH][0])
parser.add_argument('--amp_lip_y', type=float, default=AMP[DEMO_CH][1])
parser.add_argument('--amp_pos', type=float, default=AMP[DEMO_CH][2])
opt_parser = parser.parse_args()
root_dir = ROOT_DIR
opt_parser.root_dir = ROOT_DIR
opt_parser.dump_dir = os.path.join(root_dir, 'dump')
opt_parser.ckpt_dir = os.path.join(root_dir, 'ckpt', opt_parser.name)
try_mkdir(opt_parser.ckpt_dir)
opt_parser.log_dir = os.path.join(root_dir, 'log')
# make directory for nn outputs
try_mkdir(opt_parser.dump_dir.replace('dump','nn_result'))
try_mkdir(os.path.join(opt_parser.dump_dir.replace('dump', 'nn_result'), opt_parser.name))
model = Audio2landmark_model(opt_parser)
model.train()