-
Notifications
You must be signed in to change notification settings - Fork 0
/
evaluator.py
172 lines (123 loc) · 5.45 KB
/
evaluator.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
import os
import numpy as np
import matplotlib.pyplot as plt
import paddle
from models.networks import *
from misc.metric_tool import ConfuseMatrixMeter
from misc.logger_tool import Logger
from utils import de_norm
import utils
# Decide which device we want to run on
# torch.cuda.current_device()
# device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
class CDEvaluator():
def __init__(self, args, dataloader):
self.dataloader = dataloader
self.n_class = args.n_class
# define G
self.net_G = define_G(args=args, gpu_ids=args.gpu_ids)
# self.device = torch.device("cuda:%s" % args.gpu_ids[0] if torch.cuda.is_available() and len(args.gpu_ids)>0
# else "cpu")
# print(self.device)
# define some other vars to record the training states
self.running_metric = ConfuseMatrixMeter(n_class=self.n_class)
# define logger file
logger_path = os.path.join(args.checkpoint_dir, 'log_test.txt')
self.logger = Logger(logger_path)
self.logger.write_dict_str(args.__dict__)
# training log
self.epoch_acc = 0
self.best_val_acc = 0.0
self.best_epoch_id = 0
self.steps_per_epoch = len(dataloader)
self.G_pred = None
self.pred_vis = None
self.batch = None
self.is_training = False
self.batch_id = 0
self.epoch_id = 0
self.checkpoint_dir = args.checkpoint_dir
self.vis_dir = args.vis_dir
# check and create model dir
if os.path.exists(self.checkpoint_dir) is False:
os.mkdir(self.checkpoint_dir)
if os.path.exists(self.vis_dir) is False:
os.mkdir(self.vis_dir)
def _load_checkpoint(self, checkpoint_name='best_ckpt.pt'):
if os.path.exists(os.path.join(self.checkpoint_dir, checkpoint_name)):
self.logger.write('loading last checkpoint...\n')
# load the entire checkpoint
checkpoint = paddle.load(os.path.join(self.checkpoint_dir, checkpoint_name), map_location=self.device)
self.net_G.load_state_dict(checkpoint['model_G_state_dict'])
# self.net_G.to(self.device)
# update some other states
self.best_val_acc = checkpoint['best_val_acc']
self.best_epoch_id = checkpoint['best_epoch_id']
self.logger.write('Eval Historical_best_acc = %.4f (at epoch %d)\n' %
(self.best_val_acc, self.best_epoch_id))
self.logger.write('\n')
else:
raise FileNotFoundError('no such checkpoint %s' % checkpoint_name)
def _visualize_pred(self):
pred = paddle.argmax(self.G_pred, axis=1, keepdim=True)
pred_vis = pred * 255
return pred_vis
def _update_metric(self):
"""
update metric
"""
target = self.batch['L'].to(self.device).detach()
G_pred = self.G_pred.detach()
G_pred = paddle.argmax(G_pred, axis=1)
current_score = self.running_metric.update_cm(pr=G_pred.cpu().numpy(), gt=target.cpu().numpy())
return current_score
def _collect_running_batch_states(self):
running_acc = self._update_metric()
m = len(self.dataloader)
if np.mod(self.batch_id, 100) == 1:
message = 'Is_training: %s. [%d,%d], running_mf1: %.5f\n' %\
(self.is_training, self.batch_id, m, running_acc)
self.logger.write(message)
if np.mod(self.batch_id, 100) == 1:
vis_input = utils.make_numpy_grid(de_norm(self.batch['A']))
vis_input2 = utils.make_numpy_grid(de_norm(self.batch['B']))
vis_pred = utils.make_numpy_grid(self._visualize_pred())
vis_gt = utils.make_numpy_grid(self.batch['L'])
vis = np.concatenate([vis_input, vis_input2, vis_pred, vis_gt], axis=0)
vis = np.clip(vis, a_min=0.0, a_max=1.0)
file_name = os.path.join(
self.vis_dir, 'eval_' + str(self.batch_id)+'.jpg')
plt.imsave(file_name, vis)
def _collect_epoch_states(self):
scores_dict = self.running_metric.get_scores()
np.save(os.path.join(self.checkpoint_dir, 'scores_dict.npy'), scores_dict)
self.epoch_acc = scores_dict['mf1']
with open(os.path.join(self.checkpoint_dir, '%s.txt' % (self.epoch_acc)),
mode='a') as file:
pass
message = ''
for k, v in scores_dict.items():
message += '%s: %.5f ' % (k, v)
self.logger.write('%s\n' % message) # save the message
self.logger.write('\n')
def _clear_cache(self):
self.running_metric.clear()
def _forward_pass(self, batch):
self.batch = batch
img_in1 = batch['A'].to(self.device)
img_in2 = batch['B'].to(self.device)
self.G_pred = self.net_G(img_in1, img_in2)[-1]
def eval_models(self,checkpoint_name='best_ckpt.pt'):
self._load_checkpoint(checkpoint_name)
################## Eval ##################
##########################################
self.logger.write('Begin evaluation...\n')
self._clear_cache()
self.is_training = False
self.net_G.eval()
# Iterate over data.
for self.batch_id, batch in enumerate(self.dataloader, 0):
with paddle.no_grad():
self._forward_pass(batch)
self._collect_running_batch_states()
self._collect_epoch_states()