-
Notifications
You must be signed in to change notification settings - Fork 8
/
opt.py
221 lines (197 loc) · 11.5 KB
/
opt.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
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
213
214
215
216
217
218
219
220
'''
Parse arguments for training. act as config
'''
import os
import os.path as osp
import configargparse
import utils.utils as ut
def parseArgs(if_redef=True):
parser = configargparse .ArgumentParser(formatter_class=configargparse .ArgumentDefaultsHelpFormatter)
# -- env settings
parser.add_argument('--ds_fd', default='/scratch/liu.shu/datasets',
help='dataset directionry') # for discovery
parser.add_argument('--output_dir', default='output', help='default output dirs') # code local.
parser.add('--modelConf', default='config/HRpose.conf', is_config_file=True, help='Path to config file')
parser.add_argument('--ifb_debug', action='store_true')
parser.add_argument('--suffix_ptn_train', default='{model}')
parser.add_argument('--suffix_exp_train', default='ts1', help='the manually given suffix for specific test')
parser.add_argument('--suffix_ptn_test', default='{testset}-{SLP_set}',
help='the suffix pattern to form name. Change accordingly to your exp (such as ds and methods ablation)')
parser.add_argument('--suffix_exp_test', default='exp', help='mannualy added suffix for test result')
# -- data setting
parser.add_argument('--prep', default='SLP_A2J', help='data preparation method')
parser.add_argument('--SLP_set', default='danaLab', help='[danaLab|simLab] for SLP section')
parser.add_argument('--mod_src', nargs='+', default=['IR'],
help='source modality list, can accept multiple modalities typical model [RGB|IR|depthRaw| PMarray]')
parser.add_argument('--cov_li', nargs='+', default=['uncover', 'cover1', 'cover2'], help='the cover conditions')
parser.add_argument('--fc_depth', type=float, default=50., help='the depth factor to pixel level')
parser.add_argument('--if_bb', action='store_true', help='if use bounding box')
# -- model
parser.add_argument('--model', default='HRpose', help='model name [MPPE3D|A2J|HRpose|') # the model to use
parser.add_argument('--n_layers_D', type=int, default=3,
help='descriminator layer number, for 8 bb, 2 layers are good')
parser.add_argument('--net_BB', default='res50',
help='backbone net type [res50|res101|res152], can extended to VGG different layers, not so important , add later')
parser.add_argument('--out_shp', default=(64, 64, -1), type=int, nargs='+',
help='the output(hm) size of the network, last dim is optional for 3d case')
# -- train setting
parser.add_argument('--trainset', nargs='+', default=['SLP'],
help='give the main ds here the iter number will follow this one') # to mod later
# parser.add_argument('--if_D', default='y', help='if use discriminator, if single ds, then automatically set to n')
parser.add_argument('--sz_pch', nargs='+', default=(256, 256), type=int, help='input image size, model 288, pix2pix 256')
parser.add_argument('--end_epoch', default=100, type=int,
help='when reach this epoch, will stop. python index style, your model will be saved as epoch_tar-1, ori 25 ')
parser.add_argument('--epoch_step', default=-1, type=int,
help='mainly for time constrained system, each time only train step epoches, -1 for all')
parser.add_argument('--trainIter', default=-1, type=int,
help='train iters each epoch, -1 for whole set. For debug purpose (DBP)')
parser.add_argument('--optimizer', default='adam', help='[adam|nadam]')
parser.add_argument('--lr', default=1e-3, type=float)
parser.add_argument('--lr_policy', default='multi_step', help='[step|plateau|multi_step|cosine]')
parser.add_argument('--lr_dec_epoch', nargs='+', type=int, default=[70, 90],
help='the lr decay epoch, each time by decay factor ori 17,21') # form string sec dec17-21 if needed
parser.add_argument('--lr_dec_factor', default=0.1, type=float)
parser.add_argument('--batch_size_pGPU', default=60, type=int, help='batch size per gpu')
# test batch size 16 what is the reason,, no idea
parser.add_argument('--gpu_ids', nargs='+', default=[0], type=int, help='the ids of the gpu')
# parser.add_argument('--if_coninue', default='y', help='if continue to train')
parser.add_argument('--start_epoch', default=-1, type=int,
help='where to being the epoch, 0 for scratch otherwise all continue')
parser.add_argument('--init_type', default='xavier', help='weight initialization mode, gain 0.02 fixed in')
parser.add_argument('--n_thread', default=10, type=int, help='how many threads')
parser.add_argument('--save_step', default=1, type=int, help='how many steps to save model')
parser.add_argument('--if_pinMem', action='store_false', help='if pin memory to accelerate. Not working on windows')
parser.add_argument('--if_finalTest', default='n',
help='if run a final test and keep the result after training session')
# -- visualization
parser.add_argument('--display_id', type=int, default=-1, help='window id of the web display')
parser.add_argument('--display_server', type=str, default="http://localhost",
help='visdom server of the web display')
parser.add_argument('--display_env', type=str, default='main',
help='visdom display environment name (default is "main")')
parser.add_argument('--display_port', type=int, default=8097, help='visdom port of the web display')
parser.add_argument('--use_html', action='store_true', help='if use html')
parser.add_argument('--display_winsize', type=int, default=256, help='display window size for both visdom and HTML')
parser.add_argument('--display_ncols', type=int, default=3,
help='if positive, display all images in a single visdom web panel with certain number of images per row.')
parser.add_argument('--update_html_freq', type=int, default=10,
help='frequency of saving training results to html, def 1000 ')
parser.add_argument('--print_freq', type=int, default=10,
help='frequency of showing training results on console, def 100')
parser.add_argument('--no_html', action='store_true',
help='do not save intermediate training results to [opt.checkpoints_dir]/[opt.name]/web/')
# -- test setting
parser.add_argument('--if_test', action='store_true')
parser.add_argument('--testset', default='SLP',
help='testset, usually single [Human3dM|ScanAva|MuPoTS|SURREAL]')
parser.add_argument('--testIter', type=int, default=-1,
help='test iterations final and epoch test, -1 for all, DBP')
parser.add_argument('--yn_flipTest', default='y')
parser.add_argument('--if_gtRtTest', default='y', help='if use gt distance for root')
parser.add_argument('--if_adj', default='y', help='if adjust the root location to adapt different dataset')
parser.add_argument('--testImg', default=None,
help='if indicate image, test will show the skeleton and 2d images of it')
parser.add_argument('--bone_type', default='A2J',
help='choose the type of joints to be test against [model|scanava|h36m|cmJoints], what to be trained')
parser.add_argument('--if_loadPreds', default='n',
help='if load preds in test func, already saved data to accelerate')
parser.add_argument('--if_test_ckpt', default='n', help='if check intermediate checkpoint')
parser.add_argument('--svVis_step', default=1, type=int, help='step to save visuals')
parser.add_argument('--test_par', default='test',
help='the exact test portion, could be [testInLoop|test|train|all], can use the model to test on train set or test set') # I just save default first
opts, _ = parser.parse_known_args() # all cmd infor
return opts
def aug_opts(opts):
# base on given opts, add necessary informations
opts.input_shape = opts.sz_pch[::-1] # tuple size
# to update ---
opts.depth_dim = opts.input_shape[0] // 4 # save as output shape, df 64. similar to intergral
opts.bbox_3d_shape = (2000, 2000, 2000) # depth, height, width,
opts.pixel_mean = (0.485, 0.456, 0.406) # perhaps for RGB normalization after divide by 255
opts.pixel_std = (0.229, 0.224, 0.225) # for h36m version remove later
opts.SLP_fd = os.path.join(opts.ds_fd, 'SLP', opts.SLP_set) # SLP folder [danaLab|simLab]
covStr = ''
if 'uncover' in opts.cov_li:
covStr += 'u'
if 'cover1' in opts.cov_li:
covStr += '1'
if 'cover2' in opts.cov_li:
covStr += '2'
modStr = '-'.join(opts.mod_src)
input_nc = 0
for mod in opts.mod_src:
if 'RGB' in mod:
input_nc+=3
else:
input_nc += 1
opts.input_nc = input_nc
opts.clipMode = '01' # for save image purpose
if not os.path.isabs(opts.output_dir):
opts.output_dir = os.path.abspath(opts.output_dir)
nmT = '-'.join(opts.trainset) # init
dct_opt = vars(opts)
# set tne naming needed attirbutes
suffix_train = (opts.suffix_ptn_train.format(
**vars(opts))) if opts.suffix_ptn_train != '' else '' # std pattern
nmT = '_'.join([nmT, modStr, covStr, suffix_train, opts.suffix_exp_train]) # ds+ ptn_suffix+ exp_suffix
opts.name = nmT # current experiment name
opts.exp_dir = osp.join(opts.output_dir, nmT)
opts.model_dir = osp.join(opts.exp_dir, 'model_dump')
opts.vis_dir = osp.join(opts.exp_dir, 'vis', opts.test_par) #during train vis/[train|test]
opts.log_dir = osp.join(opts.exp_dir, 'log')
opts.rst_dir = osp.join(opts.exp_dir, 'result')
opts.num_gpus = len(opts.gpu_ids)
opts.web_dir = osp.join(opts.exp_dir, 'web')
vis_sub = opts.testset
if 'SLP' == opts.testset:
vis_sub = '_'.join(['SLP', opts.SLP_set]) # SLP add split to it
opts.vis_test_dir = osp.join(opts.vis_dir, vis_sub) # specific test dataset
opts.batch_size = opts.batch_size_pGPU * opts.num_gpus # the actual batch size
yn_dict = {'y': True, 'n': False}
opts.if_flipTest = yn_dict[opts.yn_flipTest]
opts.use_gt_info = yn_dict[opts.if_gtRtTest]
# test name needs start_epoch
sfx_test = (opts.suffix_ptn_test.format(**vars(opts))) if opts.suffix_ptn_test != '' else ''
opts.nmTest = '_'.join((sfx_test, opts.suffix_exp_test))
# otherwise, do nothing use the current start_epoch
return opts
def print_options(opt, if_sv=False):
"""Print and save options
It will print both current options and default values(if different).
It will save options into a text file / [checkpoints_dir] / opt.txt
"""
message = ''
message += '----------------- Options ---------------\n'
for k, v in sorted(vars(opt).items()):
comment = ''
# default = self.parser.get_default(k)
# if v != default:
# comment = '\t[default: %s]' % str(default)
message += '{:>25}: {:<30}{}\n'.format(str(k), str(v), comment)
message += '----------------- End -------------------'
print(message)
# save to the disk
if if_sv:
ut.make_folder(opt.exp_dir) # all option will mk dirs # saved to json file in set_env
file_name = os.path.join(opt.exp_dir, 'opts.txt'.format(
opt.start_epoch)) #
with open(file_name, 'wt') as opt_file:
opt_file.write(message)
opt_file.write('\n')
# opts = parseArgs()
def set_env(opts): # to be changed accordingly for rst fd
# set sys paths
# sys.path.insert(0, 'common') # not using commong
from utils.utils import add_pypath, make_folder
add_pypath(osp.join('data')) # actually we can use ds directly
for i in range(len(opts.trainset)):
add_pypath(osp.join('data', opts.trainset[i]))
# if opts.cocoapi_dir:
# add_pypath(opts.cocoapi_dir) # add coco dir to it
add_pypath(osp.join('data', opts.testset))
# add folders
make_folder(opts.model_dir)
make_folder(opts.vis_dir)
make_folder(opts.log_dir)
make_folder(opts.rst_dir)
make_folder(opts.web_dir)