-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
148 lines (115 loc) · 5.99 KB
/
test.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
import os
import sys
import math
import json
import torch
import shutil
import logging
import argparse
import numpy as np
from tqdm import tqdm
from medpy import metric
import torch.nn.functional as F
from collections import namedtuple
from torch.utils.data import DataLoader
from networks.unet import UNet
from val import test_single_case
from utils.visualize import visualize
from dataloaders.base_dataset import BaseDataset
eval_metrics = [metric.binary.dc, metric.binary.hd95, metric.binary.precision, metric.binary.recall]
n_eval = len(eval_metrics)
param = None
def test_all_case(net, param, testloader, gpu_id, stride_xy=64, stride_z=64, draw_ddm_im=False):
logging.info(param)
if os.path.exists(os.path.join(param.path.path_to_test, "test_log.txt")):
os.remove(os.path.join(param.path.path_to_test, "test_log.txt"))
total_metric_c = np.zeros((param.dataset.n_coarse - 1, n_eval))
total_metric_f = np.zeros((param.dataset.n_fine, n_eval))
all_total_metric_f = np.zeros((len(testloader), param.dataset.n_fine - 1, n_eval))
n_images = len(testloader)
tsne_index = 0 if draw_ddm_im else -1
print("Testing begin")
with open(os.path.join(param.path.path_to_dataset, 'test.list'), 'r') as fp:
image_ids = fp.readlines()
logging.info('test metrics:\t' + '\t'.join([method.__name__ for method in eval_metrics]) + '\n')
for case_index, sampled_batch in enumerate(tqdm(testloader)):
ids = image_ids[case_index].strip()
metric_c, metric_f, feat_map = test_single_case(
net, param, sampled_batch, stride_xy, stride_z, gpu_id=gpu_id, save_pred=False, ids=ids
)
if case_index == tsne_index:
if not visualize(
feat_map, sampled_batch['fine'][0], 0, 'tsne', param,
os.path.join(param.path.path_to_test, f'tsne_{param.exp.exp_name}.eps'),
legend=param.dataset.legend, n_components=2
):
tsne_index += 1
for c in range(param.dataset.n_coarse - 1):
total_metric_c[c] += metric_c[c]
logging.debug(f'{ids}\t' + '\t'.join([f"{metric_c[c, k]:.3f}" for k in range(n_eval)]))
for f in range(param.dataset.n_fine - 1):
total_metric_f[f] += metric_f[f]
all_total_metric_f[case_index] = metric_f[:-1]
logging.debug(f'{ids}\t' + '\t'.join([f"{metric_f[f, k]:.3f}" for k in range(n_eval)]))
logging.debug(f'avg fine for {ids}\t' + '\t'.join([f"{metric_f[-1, k]:.3f}" for k in range(n_eval)]))
for i in range(1, param.dataset.n_coarse):
log = f'mean of superclass {i}:\t' + '\t'.join([f"{_:.3f}" for _ in (total_metric_c[i-1] / n_images)])
logging.info(log)
for i in range(1, param.dataset.n_fine):
log = f'mean of subclass {i}:\t' + '\t'.join([f"{_:.3f}" for _ in total_metric_f[i-1] / n_images])
logging.info(log)
log = f'std of subclass {i}:\t' + '\t'.join([f"{_:.3f}" for _ in np.std(all_total_metric_f[:, i-1], axis=0)])
logging.info(log)
mean_f = [total_metric_f[:, i].sum() / (param.dataset.n_fine - 1) for i in range(len(eval_metrics))]
logging.info(f'mean of subclasses:\t' + '\t'.join([f'{i / n_images:.3f}' for i in mean_f]))
# logging.info(f'std of all subclasses: {np.std(all_total_metric_f[:, i-1]):.5f}')
total_metric_f[-1] = mean_f
return total_metric_c / n_images, total_metric_f / n_images
def calculate_metric_percase(pred, gt):
ret = np.zeros((len(eval_metrics),))
if pred.sum() > 0 and gt.sum() > 0:
for i, met in enumerate(eval_metrics):
ret[i] = met(pred, gt)
return ret
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('-m', '--model', type=str, choices=['branched', 'unet'], default='unet', help='network type for selected trained model')
parser.add_argument('-p', '--path', type=str, default='/nas/dailinrui/SSL4MIS/model_final/prostate/unet24', help='root dir of trained folder')
parser.add_argument('-g', '--gpu', type=int, default=5, help='gpu on which to test model')
args = parser.parse_args()
with open(os.path.join(args.path, 'param.json'), 'r') as fp:
d = json.load(fp)
d1 = d['dataset']
d2 = d['exp']
d3 = d['path']
d4 = d['network']
P = namedtuple('P', ['dataset', 'exp', 'path', 'network'])
param = P(dataset=namedtuple('dataset', d1.keys())(*d1.values()),
exp=namedtuple('exp', d2.keys())(*d2.values()),
path=namedtuple('path', d3.keys())(*d3.values()),
network=namedtuple('network', d4.keys())(*d4.values()))
logging.basicConfig(
level=logging.INFO, format='%(asctime)s [%(levelname)-5s] %(message)s',
datefmt='%H:%M:%S',
handlers=[logging.FileHandler(os.path.join(param.path.path_to_test, "test_log.txt"), mode='w'),
logging.StreamHandler(sys.stdout)]
)
num_classes = (param.dataset.n_coarse, param.dataset.n_fine)
test_save_path = param.path.path_to_test
if args.model == 'branched':
if param.dataset.n_coarse > 2:
from networks.multiplebranchedunet import UNetMultiBranchNetwork
net = UNetMultiBranchNetwork(param).cuda(args.gpu)
elif param.dataset.n_coarse == 2:
from networks.singlybranchedunet import UNetSingleBranchNetwork
net = UNetSingleBranchNetwork(param).cuda(args.gpu)
elif args.model == 'unet':
net = UNet(param).cuda(args.gpu)
save_mode_path = os.path.join(param.path.path_to_model, '{}_best_model.pth'.format(param.exp.exp_name))
state_dicts = torch.load(save_mode_path, map_location='cpu')
net.load_state_dict(state_dicts['model_state_dict'])
print("init weight from {}".format(save_mode_path))
net.eval()
db_test = BaseDataset(param, split='test')
testloader = DataLoader(db_test, num_workers=1, batch_size=1)
test_all_case(net, param, testloader, stride_xy=64, stride_z=64, gpu_id=args.gpu)