-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmodel_tester.py
210 lines (144 loc) · 6.7 KB
/
model_tester.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
import pandas as pd
import numpy as np
import torch
import cv2
import csv
import os
from image_dataset import ImageDataset
class ModelTester():
def __init__(self, data_transform, logger):
self.model = None
self.data_transform = data_transform
self.logger = logger
self._load_device()
self._load_trained_model()
self._load_labels_dict()
self.results = {}
def _load_device(self):
self.device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
def _load_trained_model(self):
model_file = self.logger.config_dict['TRAINED_MODEL']
self.logger.log("Start loading model from {}...".format(model_file))
self.model = torch.load(self.logger.get_model_file(model_file))
self.model.to(self.device)
self.logger.log("Finish loading model", show_time = True)
self.model_type = model_file.split('_')[0]
def _load_labels_dict(self):
df = pd.read_csv(self.logger.get_data_file(self.logger.config_dict['LABELS_FILE']))
labels = df['ClassId'].values
paths = [os.path.basename(path) for path in df['Path'].values]
self.labels_dict = dict(zip(paths, labels))
def _load_orig_test_data(self):
path_in_data_folder = self.logger.config_dict['TEST_FOLDER']
self._load_test_data(path_in_data_folder)
def _load_aug_test_data(self, folder_name):
path_in_data_folder = os.path.join(self.logger.config_dict['AUG_TEST_FOLDER'],
folder_name)
self._load_test_data(path_in_data_folder)
def _load_test_data(self, path_in_data_folder):
data, labels, fnames = [], [], []
full_path = os.path.join(self.logger.data_folder, path_in_data_folder)
enc_path = os.fsencode(full_path)
for file in os.listdir(enc_path):
filename = os.fsdecode(file)
if filename.endswith(".png"):
data.append(cv2.imread(self.logger.get_data_file(filename, path_in_data_folder)))
labels.append(self.labels_dict[filename])
fnames.append(filename)
data_dict = {'X': data, 'y': labels}
self.crt_dataset = ImageDataset(data_dict, self.logger, transform = self.data_transform,
fnames = fnames)
self.crt_folder = os.path.basename(path_in_data_folder)
self.crt_size = len(fnames)
def _get_mean_confidence(self):
data_loader = torch.utils.data.DataLoader(self.crt_dataset, batch_size = 256, shuffle = False)
self.model.eval()
pred_mask, probs_list = [], []
for batch in data_loader:
inputs = batch['image']
labels = batch['label']
fnames = batch['fname']
inputs_pt = inputs.to(self.device)
labels_pt = labels.to(self.device)
outputs_pt = self.model(inputs_pt)
outputs_pt = torch.nn.Softmax(dim = 0)(outputs_pt)
probs, preds = torch.max(outputs_pt, dim = 1)
#probs, preds = probs.numpy(), preds.numpy()
pred_mask += (labels.cpu().numpy() == preds.cpu().numpy()).tolist()
probs_list += probs.detach().cpu().numpy().tolist()
wrong_img_idxs = np.where(np.array(pred_mask) == False)[0].tolist()
mask = np.ones(len(probs_list), dtype = bool)
mask[wrong_img_idxs] = False
correct_probs = np.array(probs_list)[mask]
wrong_probs = np.array(probs_list)[wrong_img_idxs]
self.logger.log("Mean confidence for correct images: {:.2f}".format(
correct_probs.mean()), tabs = 1)
self.logger.log("Mean confidence for wrong images: {:.2f}".format(
wrong_probs.mean()), tabs = 1)
def _run_prediction(self):
data_loader = torch.utils.data.DataLoader(self.crt_dataset, batch_size = 256, shuffle = False)
self.model.eval()
pred_mask, img_names, img_preds = [], [], []
for batch in data_loader:
inputs = batch['image']
labels = batch['label']
fnames = batch['fname']
inputs_pt = inputs.to(self.device)
labels_pt = labels.to(self.device)
outputs_pt = self.model(inputs_pt)
probs, preds = torch.max(outputs_pt, dim = 1)
#probs, preds = probs.numpy(), preds.numpy()
pred_mask += (labels.cpu().numpy() == preds.cpu().numpy()).tolist()
img_names += fnames
img_preds += preds.cpu().numpy().tolist()
wrong_img_idxs = np.where(np.array(pred_mask) == False)[0].tolist()
wrong_img_names = np.array(img_names)[wrong_img_idxs].tolist()
crt_results = []
for i, (img, pred) in enumerate(zip(img_names, img_preds)):
if i in wrong_img_idxs:
crt_results.append([img, self.labels_dict[img], str(pred)])
if img not in self.results:
self.results[img] = [self.labels_dict[img], str(pred)]
else:
self.results[img].append(str(pred))
accuracy = sum(pred_mask) / self.crt_size
self.logger.log("Accuracy at test: {:.2f}".format(accuracy), tabs = 1)
self.logger.log("Number of wrong images: {}".format(len(wrong_img_idxs)), tabs = 1)
filename_wrong_imgs = self.model_type + "_" + self.crt_folder + "_wrong" + ".csv"
self.logger.log("Save wrong images filenames to {}".format(filename_wrong_imgs))
crt_results_df = pd.DataFrame(crt_results, columns = ["Name", "Orig_Label", "Pred_Label"])
crt_results_df.to_csv(self.logger.get_output_file(filename_wrong_imgs), index = False)
def _run_test_on_aug(self, folders_list):
for folder in folders_list:
self.logger.log("Test on {}".format(folder))
self._load_aug_test_data(folder)
self._get_mean_confidence()
#self._run_prediction()
def _run_test_on_orig(self):
self.logger.log("Test on original")
self._load_orig_test_data()
self._get_mean_confidence()
#self._run_prediction()
def _save_results(self):
rows = [np.concatenate(
[[key], values]).tolist() for key, values in self.results.items()]
self.rows = rows
results_df = pd.DataFrame(rows, columns = self.res_columns)
results_filename = self.model_type + "_results.csv"
results_df.to_csv(self.logger.get_output_file(results_filename), index = False)
def run_tests(self):
self._run_test_on_orig()
self.res_columns = ["Name", "Orig_Label", "Pred_Label"]
blurred_folders = ["Blurred_" + str(i) for i in range(5, 25, 5)]
self._run_test_on_aug(blurred_folders)
self.res_columns += ["Pred_" + folder + "_Label" for folder in blurred_folders]
dark_folders = ["Dark_" + str(i) for i in range(1, 4)]
self._run_test_on_aug(dark_folders)
self.res_columns += ["Pred_" + folder + "_Label" for folder in dark_folders]
bright_folders = ["Bright_" + str(i) for i in range(1, 8)]
self._run_test_on_aug(bright_folders)
self.res_columns += ["Pred_" + folder + "_Label" for folder in bright_folders]
occlud_folders = ["Occl_" + str(i) for i in range(5, 30, 5)]
self._run_test_on_aug(occlud_folders)
self.res_columns += ["Pred_" + folder + "_Label" for folder in occlud_folders]
#self._save_results()