-
Notifications
You must be signed in to change notification settings - Fork 0
/
test.py
135 lines (116 loc) · 4.84 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
#!/usr/bin/env python
import argparse
import os
import numpy as np
import torch
from PIL import Image
from torchvision import transforms
from torchvision.datasets import ImageFolder
from tqdm import tqdm
from datasets import Fitzpatrick17kDataset, ISICDataset, DDIDataset
from models import Generator
from models import DeepDermClassifier
from models import ModelDermClassifier
from models import ScanomaClassifier
from models import SSCDClassifier
from models import SIIMISICClassifier
from evaluate_classifiers import CLASSIFIER_CLASS, DATASET_CLASS
# offset the generated images from the original image by IM_OFFSET pixels
IMG_OFFSET = 50
NUM_CLASSES = 10
DEVICE = 'cuda'
DEFAULT_DATASET = ISICDataset
DEFAULT_CLASSIFIER = DeepDermClassifier
def main():
parser = argparse.ArgumentParser()
parser.add_argument("--checkpoint_path", type=str, default="checkpoint.pth")
parser.add_argument("--dataset", type=str, choices=["f17k", "isic", "ddi", "from_file"], default="from_file")
parser.add_argument("--classifier", type=str, choices=["deepderm", "modelderm", "scanoma", "sscd", "siimisic", "from_file"], default="from_file")
parser.add_argument("--output", type=str, default="out")
parser.add_argument("--max_images", type=int, default=40)
parser.add_argument("--batch_size", type=int, default=4)
args = parser.parse_args()
outdir = args.output
if not os.path.exists(outdir):
print(f"...Creating output directory {outdir}")
os.mkdir(outdir)
checkpoint_path = args.checkpoint_path
if args.dataset == "from_file":
dataset_class = DEFAULT_DATASET
else:
dataset_class = DATASET_CLASS[args.dataset]
if args.classifier == "from_file":
classifier_class = DEFAULT_CLASSIFIER
else:
classifier_class = CLASSIFIER_CLASS[args.classifier]
# Load classifier model
classifier = classifier_class()
positive_index = classifier.positive_index
classifier.eval()
im_size = classifier.image_size
# Load generator model
generator = Generator(im_size=im_size)
checkpoint = torch.load(checkpoint_path)
generator.load_state_dict(checkpoint['generator'])
normalize = transforms.Normalize(mean=0.5,
std=0.5)
transform = transforms.Compose([
transforms.Resize(im_size),
transforms.CenterCrop(im_size),
transforms.ToTensor(),
normalize])
dataset = dataset_class(transform=transform)
dataloader = torch.utils.data.DataLoader(
dataset,
batch_size=args.batch_size,
shuffle=False,
drop_last=True,
num_workers=2)
generator.to(DEVICE)
classifier.to(DEVICE)
# for SIIM-ISIC
try: classifier.enable_augment()
except AttributeError: pass
for ibatch, batch in enumerate(tqdm(dataloader)):
if ibatch*args.batch_size > args.max_images: break
img, label = batch
img = img.to(DEVICE)
targets_min = torch.zeros(img.shape[0], dtype=torch.long).to(DEVICE)
targets_max = (NUM_CLASSES-1)*torch.ones(img.shape[0], dtype=torch.long).to(DEVICE)
with torch.no_grad():
# transform images
img_min, _ = generator(img, targets_min)
img_max, _ = generator(img, targets_max)
# check classifier predictions
pred_orig = classifier(img)[:,positive_index]
pred_min = classifier(img_min)[:,positive_index]
pred_max = classifier(img_max)[:,positive_index]
# save images separately
for i_img in range(img.shape[0]):
index = i_img + ibatch*args.batch_size
orig = img[i_img].detach().cpu().numpy()
min_ = img_min[i_img].detach().cpu().numpy()
max_ = img_max[i_img].detach().cpu().numpy()
orig_label = label[i_img]
pred_orig_ = pred_orig[i_img].detach().cpu().numpy()
pred_min_ = pred_min[i_img].detach().cpu().numpy()
pred_max_ = pred_max[i_img].detach().cpu().numpy()
full = np.ones((im_size, im_size*3+IMG_OFFSET, 3))
full[:,:im_size,:] = orig.swapaxes(0,1).swapaxes(1,2)
full[:,IMG_OFFSET+im_size:IMG_OFFSET+2*im_size,:] = min_.swapaxes(0,1).swapaxes(1,2)
full[:,IMG_OFFSET+2*im_size:,:] = max_.swapaxes(0,1).swapaxes(1,2)
full *= 0.5
full += 0.5
full *= 255
full = np.require(full, dtype=np.uint8)
out_img = Image.fromarray(full)
out_img.save(os.path.join(outdir,
"{:05d}_{:d}_{:.03f}_{:.03f}_{:.03f}.png"\
.format(index,
int(orig_label),
pred_orig_,
pred_min_,
pred_max_))
)
if __name__ == "__main__":
main()