-
Notifications
You must be signed in to change notification settings - Fork 1
/
predict_heatmap.py
241 lines (206 loc) · 9.17 KB
/
predict_heatmap.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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
##############################
# code for predicting anomaly detection heatmaps of images
# using pluralistic image completion.
##############################
### IMPORTS
import os
from argparse import ArgumentParser
from utils import *
from modules import *
from heatmapping import *
from eval import *
# torch
import torch
import torchvision
import torchvision.transforms as transforms
import torch.nn.functional as F
# other libs
from datetime import datetime
import random
from random import sample
import matplotlib
import matplotlib.pyplot as plt
from tqdm import tqdm
import numpy as np
from time import time
parser = ArgumentParser()
parser.add_argument('--config', type=str, default='configs/config.yaml',
help="configuration")
parser.add_argument('--seed', type=int, default=1337,
help="manual random seed")
parser.add_argument('--checkpoint_dir', type=str,
help="path to saved inpainter model checkpoint directory")
parser.add_argument('--checkpoint_iter', type=int,
help="iteration number of saved model checkpoint")
def main():
args = parser.parse_args()
config = get_config(args.config)
############################################################
### (1) GPU setup
############################################################
cuda = config['cuda']
device_check = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print('running on {}'.format(device_check))
# set which devices CUDA sees
device_ids = config['gpu_ids'] # indices of devices for models, data and otherwise
if cuda:
os.environ['CUDA_VISIBLE_DEVICES'] = ','.join(str(i) for i in device_ids)
# all devices are then indexed from this set
model_device = 0
# set random seed
seed = args.seed
random.seed(seed)
torch.manual_seed(seed)
if cuda:
torch.cuda.manual_seed_all(seed)
############################################################
### (2) model setup
############################################################
# choose model and dataset we'll be working with
model_type = 'dropout'
dataset_name = config['dataset_name']
PIL_img_format = 'L' if (config["test"]["patch_shape"][-1] == 1) else 'RGB'
completion_img_size = config["test"]["patch_shape"][0]
# ^ side size of image to be completed (may be just part of larger heatmap image)
# load utils
normalize_img = load_img_normalizer(model_type)
# hyperparameter and checkpoint setup
hyperparams = {
'dropout' : {
'p_dropout' : config["test"]["droprate"]
},
}
checkpoints = {
'dropout' : {
'gen' : args.checkpoint_dir,
'dis' : args.checkpoint_dir,
'iter' : args.checkpoint_iter
},
}
# load inpainter and completion feature extractor
inpainter = load_multi_inpainter(
model_type,
checkpoints[model_type],
hyperparams[model_type],
device_ids,
dropoutmodel_config=args.config
)
feature_extractor = load_inpainting_feature_extractor(
model_type,
checkpoints[model_type],
hyperparams[model_type],
device_ids,
dropoutmodel_config=args.config
)
# heatmapping settings
# visualization and analysis settings
save_heatmap_data = config["test"]["save_heatmap_data"]
save_heatmap_plots = config["test"]["save_heatmap_plots"]
save_progressive_heatmap = config["test"]["save_progressive_heatmap"]
log_compute_times = config["test"]["log_compute_times"]
# heatmapping parameters
mask_size = config["test"]["mask_shape"][0]
window_size = config["test"]["patch_shape"][0]
window_stride = config["test"]["patch_stride"]
heatmap_M_inpaint = config["test"]["heatmap_M_inpaint"]
heatmap_metrics = config["test"]["heatmap_metrics"]
parallel_batchsize = config["test"]["parallel_batchsize"]
# misc scoring settings
only_check_nonblack_pixels = config["test"]["only_check_nonblack_pixels"]
# logger
log_dir = 'test_logs'
if not os.path.exists(log_dir):
os.makedirs(log_dir)
logger = Logger('test', log_dir, heatmap_metrics)
dt_now = datetime.now()
############################################################
### (3) heatmap generation
############################################################
test_data_fnames = [os.path.join(config["test_data_path"], f) for f in os.listdir(config["test_data_path"])]
with torch.no_grad():
for test_data in test_data_fnames:
# 1) load image
img = pil_loader(test_data, img_format=PIL_img_format)
img = transforms.ToTensor()(img)
# don't normalize at the image level: will normalize at the patch level
img = img.unsqueeze(dim=0)
# plot bbox on img
img = img.cpu()
print(test_data)
show_images(img, custom_figsize=(10, 14))
img = img.cuda()
ignore_mask = None
if only_check_nonblack_pixels:
print('ONLY CHECKING NONBLACK PIXELS')
# mask of size image; True where there are pixels that
# we don't want to include in evaluation
ignore_mask = (normalize_img(img) == -1.).cpu()
# 2) generate heatmaps
tin = time()
heatmaps = generate_anomaly_heatmap_slidingwindow_PARALLEL(
img,
inpainter,
feature_extractor,
metrics=heatmap_metrics,
mask_size=mask_size,
window_size=window_size,
window_stride=window_stride,
M_inpaint=heatmap_M_inpaint,
heatmap_batch_size=parallel_batchsize,
heatmap_type='nonaveraged',
img_normalizer = normalize_img,
save_progressive_heatmap = save_progressive_heatmap
)
tout = time()
print('time to create heatmap = {} sec'.format(tout - tin))
# plot and save heatmap data and images
for heatmap_metric in heatmap_metrics:
# create dirs
savedir = os.path.join('heatmaps', dataset_name, model_type, dt_now.strftime("%m-%d-%Y_%H:%M:%S"))
savedir_maps = os.path.join(savedir, 'data')
savedir_plots = os.path.join(savedir, 'plots')
for path in [savedir_maps, savedir_plots]:
if not os.path.exists(path):
os.makedirs(path)
# save heatmap data
filename = test_data
filename = filename.replace('.png', '')
filename = filename.split('/')
filename = filename[-1]
filename += '_{}_{}_{}_{}_{}_{}.pt'.format(heatmap_metric,
heatmap_M_inpaint,
hyperparams['dropout']['p_dropout'],
mask_size,
window_size,
window_stride,
)
filename_map = os.path.join(savedir_maps, filename)
if save_heatmap_data:
torch.save(heatmaps[heatmap_metric], filename_map)
# plot heatmap
fig, ax = plt.subplots(figsize=(10, 14))
im = ax.imshow((heatmaps[heatmap_metric]).cpu(), cmap=plt.cm.hot, interpolation='none')
cbar = fig.colorbar(im, extend='max')
title = 'anomaly metric: {}\nM={}, p={}, mask size = {}\nwindow size = {}, window stride = {}'.format(
heatmap_metric,
heatmap_M_inpaint,
hyperparams['dropout']['p_dropout'],
mask_size,
window_size,
window_stride,
)
plt.title(title, fontsize=20)
# save heatmap plot 'data_new/test/cancer/val_DBT-P01700_DBT-S01353_lmlo_Cancer_0.png'
filename_img = filename.replace('.pt', '.png')
filename_img = os.path.join(savedir_plots, filename_img)
if save_heatmap_plots:
plt.savefig(fname=filename_img, bbox_inches = 'tight')
plt.show()
# log heatmaps on image
log_hyperparams = [window_stride, window_size, mask_size,
heatmap_M_inpaint, parallel_batchsize, 'nonaveraged',
hyperparams['dropout']['p_dropout']]
if log_compute_times:
logger.write_msg('heatmap compute time on {} GPUs = {}\n'.format(len(device_ids), tout-tin))
if __name__ == '__main__':
main()