forked from aimi-bonn/Deeplasia
-
Notifications
You must be signed in to change notification settings - Fork 0
/
predict.py
171 lines (139 loc) · 5.32 KB
/
predict.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
import os
import logging.config
import pandas as pd
import torch
import yaml
from lib.utils.log import LOG_CONFIG
logging.config.dictConfig(LOG_CONFIG)
from argparse import ArgumentParser
import pytorch_lightning as pl
from torch.utils.data import DataLoader
from lib.datasets import InferenceDataset
from lib.models import BoneAgeModel
from lib.legacy import from_checkpoint as load_legacy_model
from lib import testing
import re
import warnings
warnings.filterwarnings("ignore")
def main():
logger = logging.getLogger()
parser = create_parser()
args = parser.parse_args()
trainer = pl.Trainer.from_argparse_args(
args, checkpoint_callback=False, logger=False
)
loader = create_loader(args, logger)
if args.legacy_ckp:
model = load_legacy_model(args)
else:
model = BoneAgeModel.load_from_checkpoint(args.ckp_path)
outputs = trainer.predict(model=model, dataloaders=loader)
y = torch.concat([o["y"] for o in outputs]) if "y" in outputs[0].keys() else None
names = [
val.split("/")[-1]
for sublist in [o["image_path"] for o in outputs]
for val in sublist
]
y_hat_out = torch.concat([o["y_hat"] for o in outputs])
sex = torch.concat([o["sex"] for o in outputs])
if not args.legacy_ckp:
sex_hat = torch.concat([o["sex_hat"] for o in outputs])
# Note that the outputs are z-scores
# they need to be re-transformed (and regression-corrected) to get the real age
df = {
"image_ID": names,
"sex": sex.squeeze(),
"y_hat": y_hat_out.squeeze(),
}
if not args.legacy_ckp:
df = df | {"sex_hat": sex_hat.squeeze()}
df = pd.DataFrame(df)
if args.legacy_ckp:
df["pred_" "bone_age"] = rescale_prediction(df["y_hat"], args.ckp_path)
if y is not None:
df["y"] = y.squeeze()
df.to_csv(args.output_path, index=False)
logger.info(f"saved to {args.output_path}")
def create_parser():
parser = ArgumentParser()
parser.add_argument("--ckp_path", type=str)
parser.add_argument(
"--backbone",
type=str,
default=None,
help="CNN backbone for the model. If not provided attempted to be inferred from the ckp path",
)
parser.add_argument(
"--legacy_ckp",
action="store_true",
help="use legacy ckp format (checkpoints from before 03/22)",
)
# data set stuff (only relevant options)
parser.add_argument("--annotation_csv", type=str, default="data/annotation.csv")
parser.add_argument(
"--split_csv", type=str, default="data/splits/rsna_original.csv"
)
parser.add_argument("--split_column", type=str, default="")
parser.add_argument("--split_name", type=str, default="test")
parser.add_argument("--img_dir", type=str, default="../data/annotated/")
parser.add_argument("--mask_dirs", nargs="+", default=["../data/masks/fscnn_cos"])
parser.add_argument("--input_size", nargs="+", default=[1, 512, 512], type=int)
parser.add_argument("--image_norm_method", type=str, default="zscore")
parser.add_argument("--mask_crop_size", type=float, default=-1)
parser.add_argument("--flip", action="store_true")
parser.add_argument("--rotation_angle", type=float, default=0)
# other options
parser.add_argument("--batch_size", type=int, default=16)
parser.add_argument("--num_workers", type=int, default=8)
parser.add_argument(
"--output_path", type=str, default="output/predictions_results.csv"
)
# store activations
parser.add_argument("--store_activations", action="store_true")
# TODO
parser = pl.Trainer.add_argparse_args(parser)
return parser
def create_loader(args, logger):
if "highRes" in args.ckp_path:
args.input_size = [1, 1024, 1024]
logger.info(
"changed input size to 1024 as 'highRes' was detected in the ckp path"
)
if not args.mask_dirs[0]:
args.mask_dirs = []
logger.info(f"using masks from {args.mask_dirs}")
loader = DataLoader(
InferenceDataset(
annotation_df=args.annotation_csv,
split_df=args.split_csv,
split_column=args.split_column,
split_name=args.split_name,
img_dir=args.img_dir,
mask_dirs=args.mask_dirs,
norm_method=args.image_norm_method,
input_size=args.input_size,
mask_crop_size=args.mask_crop_size,
flip=args.flip,
rotation_angle=args.rotation_angle,
),
num_workers=args.num_workers,
batch_size=args.batch_size,
drop_last=False,
shuffle=False,
)
return loader
def rescale_prediction(y_hat, ckp_path, params_path="data/parameters.yml"):
with open(params_path, "r") as stream:
cor_params = yaml.safe_load(stream)
def cor_prediction_bias(yhat, slope, intercept):
"""corrects model predictions (yhat) for linear bias (defined by slope and intercept)"""
return yhat - (yhat * slope + intercept)
age_mean, age_sd = cor_params["age_mean"], cor_params["age_sd"]
y_hat = y_hat * age_sd + age_mean
ckp_path = ckp_path.split("/")[-1].split(".")[0]
slope = cor_params[ckp_path]["slope"]
intercept = cor_params[ckp_path]["intercept"]
y_hat = cor_prediction_bias(y_hat, slope, intercept)
return y_hat
if __name__ == "__main__":
main()