-
Notifications
You must be signed in to change notification settings - Fork 4
/
Copy pathconfig.py
77 lines (59 loc) · 2.53 KB
/
config.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
import os
from easydict import EasyDict as edict
import time
import torch
__C = edict()
cfg = __C
__C.seed = 3035 # random seed, for reproduction
# ---------------the parameters of training----------------
__C.sou_dataset = 'GCC' #
__C.model_type = 'vgg16' # choices=['ResNet50','vgg16']
__C.phase ='DA_train' # choices=['DA_train','pre_train' , 'fine_tune'])
__C.gpu_id = "0,1" # single gpu:"0"..; multi gpus:"2,3,
__C.target_dataset ='SHHB' # dataset choices = ['SHHB', 'UCF50', 'QNRF', 'MALL', 'UCSD', 'SHHA']
__C.init_weights=None
# ======== nlt learning rate setting==============
__C.step_size=10 # The number of epochs to reduce the meta learning rates
__C.gamma=0.98 # Gamma for the meta-train learning rate decay
__C.nlt_lr = 1e-5 # MCNN：1e-4 Res50:1e-5
__C.nlt_lr_decay = 1e-4 # learning rate decay rate
__C.DA_stop_epoch = 200
# ======Parameters for pretain phase===========
__C.pre_lr=1e-5 # Learning rate for pre-train phase
__C.pre_weight_decay=1e-4 # Weight decay for the optimizer during pre-train
__C.pre_gamma=0.98 # Gamma for the pre-train learning rate decay
__C.pre_max_epoch=100 # Epoch number for pre-train phase
__C.pre_step_size = 2
__C.pre_batch_size = 8
# ======Parameters for fine_tune phase===========
__C.GCC_pre_train_model = './exp/SHHB/pre/04-01_07-17_SHHB_vgg16__lr1e-05pre_train/all_ep_5_mae_87.7_mse_198.7.pth'
__C.fine_lr = 1e-6
__C.fine_weight_decay=1e-4 # Weight decay for the optimizer during fine_tune
__C.fine_step_size = 2
__C.fine_gamma=0.98 # Gamma for the pre-train learning rate decay
# ==============common parameters==================
__C.print_freq = 50 # print frequency
__C.max_epoch=250
__C.VAL_DENSE_START = 150
__C.val_freq = 1 # Before __C.VAL_DENSE_START epoches, the freq is set as __C.VAL_FREQ
# ------------------------------VIS------------------------
now = time.strftime("%m-%d_%H-%M", time.localtime())
__C.EXP_NAME = now + "_" + __C.target_dataset + "_" + __C.model_type + "_" +\
'_lr' + str(__C.pre_lr) + '_'+\
__C.phase
if __C.target_dataset == "SHHA":
__C.EXP_PATH = "./exp/SHHA"
if __C.target_dataset == "QNRF":
__C.EXP_PATH = "./exp/QNRF"
if __C.target_dataset == "SHHB":
__C.EXP_PATH = "./exp/SHHB"
if __C.target_dataset == "MALL":
__C.EXP_PATH = "./exp/MALL"
if __C.target_dataset == "UCSD":
__C.EXP_PATH = "./exp/UCSD"
if __C.target_dataset == "UCF50":
__C.EXP_PATH = "./exp/UCF50"
if __C.target_dataset == "GCC":
__C.EXP_PATH = "./exp"
if __C.target_dataset == "WE":
__C.EXP_PATH = "./exp/WE"