-
Notifications
You must be signed in to change notification settings - Fork 12
/
run_sample_coco.py
185 lines (148 loc) · 7.67 KB
/
run_sample_coco.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
import argparse
import os
import numpy as np
import os.path as osp
from misc import pyutils
import random
import torch
# def seed_torch(seed=1):
# random.seed(seed)
# os.environ['PYTHONHASHSEED'] = str(seed)
# np.random.seed(seed)
# torch.manual_seed(seed)
# torch.cuda.manual_seed(seed)
# torch.cuda.manual_seed_all(seed) # if you are using multi-GPU.
# torch.backends.cudnn.benchmark = False
# torch.backends.cudnn.deterministic = True
# torch.backends.cudnn.enabled = False
if __name__ == '__main__':
# seed_torch(seed=2)
def str2bool(v):
if isinstance(v, bool):
return v
if v.lower() in ('yes', 'true', 't', 'y', '1'):
return True
elif v.lower() in ('no', 'false', 'f', 'n', '0'):
return False
else:
raise argparse.ArgumentTypeError('Boolean value expected.')
parser = argparse.ArgumentParser()
# Environment
# parser.add_argument("--num_workers", default=os.cpu_count()//2, type=int)
parser.add_argument("--num_workers", default=12, type=int)
parser.add_argument("--mscoco_root", default='/data0/dataset/xjheng/COCO/', type=str, help="Path to MSCOCO")
parser.add_argument("--num_classes", default=80, type=int)
# Class Activation Map
parser.add_argument("--cam_network", default="net.resnet50_cam", type=str)
parser.add_argument("--feature_dim", default=2048, type=int)
parser.add_argument("--cam_crop_size", default=512, type=int)
parser.add_argument("--cam_batch_size", default=16, type=int)
parser.add_argument("--cam_num_epoches", default=5, type=int)
parser.add_argument("--cam_learning_rate", default=0.1, type=float)
parser.add_argument("--cam_weight_decay", default=1e-4, type=float)
parser.add_argument("--cam_eval_thres", default=0.1, type=float)
parser.add_argument("--cam_scales", default=(1.0, 0.5, 1.5, 2.0),
help="Multi-scale inferences")
parser.add_argument("--num_cores_eval", default=8, type=int)
# CLIMS
parser.add_argument("--clims_network", default="net.resnet50_clims", type=str)
parser.add_argument("--clims_learning_rate", default=0.00025, type=float)
parser.add_argument("--clims_num_epoches", default=15, type=int)
parser.add_argument('--hyper', default='2,14,2,0.2', type=str)
parser.add_argument('--clip', default='ViT-B/32', type=str)
# Mining Inter-pixel Relations
parser.add_argument("--conf_fg_thres", default=0.35, type=float)
parser.add_argument("--conf_bg_thres", default=0.1, type=float)
# Inter-pixel Relation Network (IRNet)
parser.add_argument("--irn_network", default="net.resnet50_irn", type=str)
parser.add_argument("--irn_crop_size", default=512, type=int)
parser.add_argument("--irn_batch_size", default=32, type=int)
parser.add_argument("--irn_num_epoches", default=3, type=int)
parser.add_argument("--irn_learning_rate", default=0.1, type=float)
parser.add_argument("--irn_weight_decay", default=1e-4, type=float)
# Random Walk Params
parser.add_argument("--beta", default=10)
parser.add_argument("--exp_times", default=8,
help="Hyper-parameter that controls the number of random walk iterations,"
"The random walk is performed 2^{exp_times}.")
parser.add_argument("--ins_seg_bg_thres", default=0.25)
parser.add_argument("--sem_seg_bg_thres", default=0.25)
# Output Path
parser.add_argument("--work_space", default="result_default5", type=str) # set your path
parser.add_argument("--log_name", default="sample_train_eval", type=str)
parser.add_argument("--cam_weights_name", default="res50_cam.pth", type=str)
parser.add_argument("--irn_weights_name", default="res50_irn.pth", type=str)
parser.add_argument("--cam_out_dir", default="cam_mask", type=str)
parser.add_argument("--ir_label_out_dir", default="ir_label", type=str)
parser.add_argument("--sem_seg_out_dir", default="sem_seg", type=str)
parser.add_argument("--ins_seg_out_dir", default="ins_seg", type=str)
parser.add_argument("--clims_weights_name", default="res50_clims", type=str)
# Step
parser.add_argument("--train_cam_pass", type=str2bool, default=False)
parser.add_argument("--train_clims_pass", type=str2bool, default=False)
parser.add_argument("--make_cam_pass", type=str2bool, default=False)
parser.add_argument("--make_clims_pass", type=str2bool, default=False)
parser.add_argument("--eval_cam_pass", type=str2bool, default=False)
parser.add_argument("--cam_to_ir_label_pass", type=str2bool, default=False)
parser.add_argument("--train_irn_pass", type=str2bool, default=False)
parser.add_argument("--make_ins_seg_pass", type=str2bool, default=False)
parser.add_argument("--eval_ins_seg_pass", type=str2bool, default=False)
parser.add_argument("--make_sem_seg_pass", type=str2bool, default=False)
parser.add_argument("--eval_sem_seg_pass", type=str2bool, default=False)
parser.add_argument("--use_distributed_train", type=str2bool, default=False)
parser.add_argument('--local_rank', default=-1, type=int,
help='NOT CHANGE! for distributed train')
parser.add_argument("--cbs_loss_thresh", type=float, default=0.285)
args = parser.parse_args()
args.log_name = osp.join(args.work_space, args.log_name)
args.cam_weights_name = osp.join(args.work_space, args.cam_weights_name)
args.irn_weights_name = osp.join(args.work_space, args.irn_weights_name)
args.cam_out_dir = osp.join(args.work_space, args.cam_out_dir)
args.ir_label_out_dir = osp.join(args.work_space, args.ir_label_out_dir)
args.sem_seg_out_dir = osp.join(args.work_space, args.sem_seg_out_dir)
args.ins_seg_out_dir = osp.join(args.work_space, args.ins_seg_out_dir)
args.clims_weights_name = osp.join(args.work_space, args.clims_weights_name)
os.makedirs(args.work_space, exist_ok=True)
os.makedirs(args.cam_out_dir, exist_ok=True)
os.makedirs(args.ir_label_out_dir, exist_ok=True)
os.makedirs(args.sem_seg_out_dir, exist_ok=True)
os.makedirs(args.ins_seg_out_dir, exist_ok=True)
pyutils.Logger(args.log_name + '.log')
print(vars(args))
if args.train_cam_pass is True:
import step_coco.train_cam
timer = pyutils.Timer('step.train_cam:')
step_coco.train_cam.run(args)
if args.train_clims_pass is True:
import step_coco.train_clims
timer = pyutils.Timer('step.train_clims:')
step_coco.train_clims.run(args)
if args.make_cam_pass is True:
import step_coco.make_cam
timer = pyutils.Timer('step.make_cam:')
step_coco.make_cam.run(args)
if args.make_clims_pass is True:
import step_coco.make_clims
timer = pyutils.Timer('step.make_clims:')
step_coco.make_clims.run(args)
if args.eval_cam_pass is True:
import step_coco.eval_cam
timer = pyutils.Timer('step.eval_cam:')
step_coco.eval_cam.run(args)
if args.cam_to_ir_label_pass is True:
import step_coco.cam_to_ir_label
timer = pyutils.Timer('step.cam_to_ir_label:')
step_coco.cam_to_ir_label.run(args)
if args.train_irn_pass is True:
import step_coco.train_irn
timer = pyutils.Timer('step.train_irn:')
step_coco.train_irn.run(args)
if args.make_sem_seg_pass is True:
import step_coco.make_sem_seg_labels
args.sem_seg_bg_thres = float(args.sem_seg_bg_thres)
timer = pyutils.Timer('step.make_sem_seg_labels:')
step_coco.make_sem_seg_labels.run(args)
if args.eval_sem_seg_pass is True:
import step_coco.eval_sem_seg
timer = pyutils.Timer('step.eval_sem_seg:')
step_coco.eval_sem_seg.run(args)