-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathtest.py
166 lines (156 loc) · 9.9 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
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
import os
import argparse
import torch
import numpy as np
from torchvision import transforms
from torch.utils.data import DataLoader
from sklearn.metrics import balanced_accuracy_score
from datasets.utils import DeepLIIFImgMaskDataset, WarwickImgMaskDataset, DeepLIIFBenchmarkDataset, \
WarwickBenchmarkDataset
from models import Generator, CustomUnet, load_optimal_hyperparameters, load_pickle_file
from postprocessing import compute_morphological_operations
from performance import dice_by_sample, dice, hausdorff, f1_m, precision_m, recall_m, obj_metric, aji_score
def main():
parser = argparse.ArgumentParser(description='Train GAN Model for IHC nuclei segmentation from spatial organization'
'prior')
parser.add_argument('data_path', type=str, default=None,
help='IHC images and/or ground truth masks root path')
parser.add_argument('method', type=str, choices=["GAN", "Unet", "Stardist", "Nuclick", "Thresholding"],
default=None, help='Choose the method to evaluate among ["GAN", "Unet", "Stardist",'
' "Nuclick", "Thresholding"]')
parser.add_argument('dataset_name', type=str, choices=["deepliif", "warwick"], default=None,
help='Choose between deepliif, bcdataset (Ki67) or warwick (HER2) dataset')
parser.add_argument('--benchmark_path', type=str, default=None,
help='Predicted masks path with benchmark methods ("Stardist", "Nuclick", "Thresholding")')
parser.add_argument('--checkpoint_path', type=str, default="./pretrained_models",
help='Best model .pt file path (should also contain G_losses_GAN.pickle for GAN method, '
'and optimal_parameters.pickle for Unet method)')
parser.add_argument('--start_epoch_checkpoint', type=int, default=250,
help='Number of epoch to skip before assessing the minimum of G_losses_GAN (for GAN method)')
parser.add_argument('--batch_size', type=int, default=None,
help='Batch size for testing')
parser.add_argument('--workers', type=int, default=5,
help='Number of worker nodes (default: 5)')
args = parser.parse_args()
if args.method in ["GAN", "Unet"] and args.checkpoint_path is None:
raise ValueError("Please provide a checkpoint path when evaluating GAN or Unet approach")
elif args.method in ["Stardist", "Nuclick", "Thresholding"] and args.benchmark_path is None:
raise ValueError("Please provide a benchmark path when evaluating existing methods")
if args.method in ["GAN", "Unet"]:
if args.dataset_name == "deepliif":
dataset_ihc_test_path = os.path.join(args.data_path, "DeepLIIF_Testing_Set")
dataset_ihc_test = DeepLIIFImgMaskDataset(dataset_ihc_test_path,
img_transform=transforms.Compose([transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5,
0.5))]),
mask_transform=transforms.ToTensor())
elif args.dataset_name == "warwick":
dataset_ihc_test_images_path = os.path.join(args.data_path, "images")
dataset_ihc_test_masks_path = os.path.join(args.data_path, "ground_truths_corrected")
dataset_ihc_test = WarwickImgMaskDataset(dataset_ihc_test_images_path, dataset_ihc_test_masks_path,
img_transform=transforms.Compose([transforms.ToTensor(),
transforms.Normalize(
(0.5, 0.5, 0.5),
(0.5, 0.5,
0.5))]),
mask_transform=transforms.ToTensor())
else:
raise NotImplementedError
else:
if args.dataset_name == "deepliif":
dataset_ihc_test_path = os.path.join(args.data_path, "DeepLIIF_Testing_Set")
dataset_ihc_test = DeepLIIFBenchmarkDataset(args.benchmark_path, dataset_ihc_test_path,
mask_transform=transforms.ToTensor(),
gt_transform=transforms.ToTensor())
elif args.dataset_name == "warwick":
dataset_ihc_test_masks_path = os.path.join(args.data_path, "ground_truths")
dataset_ihc_test = WarwickBenchmarkDataset(args.benchmark_path, dataset_ihc_test_masks_path,
mask_transform=transforms.ToTensor(),
gt_transform=transforms.ToTensor())
else:
raise NotImplementedError
print(f"Testing {args.method} method on {len(dataset_ihc_test)} data...")
if args.batch_size:
dataloader_ihc_test = DataLoader(dataset_ihc_test, batch_size=args.batch_size, shuffle=False,
num_workers=args.workers)
else:
dataloader_ihc_test = DataLoader(dataset_ihc_test, batch_size=len(dataset_ihc_test), shuffle=False,
num_workers=args.workers)
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
if args.method == "GAN":
if os.path.isfile(os.path.join(args.checkpoint_path, "G_losses_system.pickle")):
G_losses_system = load_pickle_file(os.path.join(args.checkpoint_path, "G_losses_system.pickle"))
start_idx = min(args.start_epoch_checkpoint, len(G_losses_system) - 1)
best_idx = np.argmin(G_losses_system[start_idx:]) + start_idx
checkpoint = torch.load(os.path.join(args.checkpoint_path, f"model_epoch{best_idx}.pt"))
print(f"Best model found for epoch {best_idx}.")
else:
checkpoint = torch.load(os.path.join(args.checkpoint_path, f"proposed_{args.dataset_name}.pt"))
model = Generator().to(device)
model.load_state_dict(checkpoint["generator_ihc_to_mask_state_dict"])
model.eval()
elif args.method == "Unet":
optimal_parameters = load_optimal_hyperparameters(os.path.join(args.checkpoint_path,
"optimal_parameters.pickle"))
ngf = optimal_parameters["ngf"]
dropout_value = optimal_parameters["dropout_value"]
checkpoint = torch.load(os.path.join(args.checkpoint_path, "best_model.pt"))
model = CustomUnet(ngf=ngf, dropout_value=dropout_value).to(device)
model.load_state_dict(checkpoint)
model.eval()
accuracies = []
f1_scores = []
precisions = []
recalls = []
dice_scores = []
dice_object = []
aji_scores = []
hausdorff_scores = []
already_labelled = args.method == "Nuclick"
watershed = args.method not in ["Stardist", "Nuclick"]
with torch.no_grad():
for data, target in dataloader_ihc_test:
if args.method in ["GAN", "Unet"]:
data = data.to(device)
pred = model(data)
pred = (pred > 0.5).float().cpu()
all_pred = []
for img in pred:
img_postprocessed = compute_morphological_operations(img.squeeze().numpy(),
erosion=args.dataset_name == "warwick")
all_pred.append(torch.tensor(img_postprocessed).unsqueeze().float())
pred = torch.stack(all_pred)
pred_obj = pred
elif args.method == "Thresholding":
pred = data
all_pred = []
for img in pred:
img_postprocessed = compute_morphological_operations(img.squeeze().numpy(),
erosion=args.dataset_name == "warwick")
all_pred.append(torch.tensor(img_postprocessed).unsqueeze().float())
pred = torch.stack(all_pred)
pred_obj = pred
elif args.method == "Nuclick":
pred = (data > 0.).float()
pred_obj = data
else:
pred = data
pred_obj = data
accuracies.append(
np.mean([balanced_accuracy_score(target[i, ...].flatten(), pred[i, ...].flatten()) for i in
range(target.shape[0])]))
f1_scores.append(f1_m(pred, target).item())
precisions.append(precision_m(pred, target).item())
recalls.append(recall_m(pred, target).item())
dice_scores.append(dice_by_sample(pred, target).item())
dice_object.append(obj_metric(pred_obj, target, dice, watershed, already_labelled))
hausdorff_scores.append(obj_metric(pred_obj, target, hausdorff, watershed, already_labelled))
aji_scores.append(aji_score(pred_obj, target, watershed, already_labelled))
print(f"Averaged test : acc = {np.mean(accuracies)}, Dice = {np.mean(dice_scores)}, "
f"Dice object = {np.mean(dice_object)} F1 score = {np.mean(f1_scores)}, "
f"precision = {np.mean(precisions)}, recall = {np.mean(recalls)},"
f"AJI = {np.mean(aji_scores)}, Hausdorff = {np.mean(hausdorff_scores)}")
if __name__ == '__main__':
main()