File size: 8,325 Bytes
599a397
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) MONAI Consortium
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#     http://www.apache.org/licenses/LICENSE-2.0
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.

from __future__ import annotations

import argparse
import json
import logging
import os
from pathlib import Path

import monai
import nibabel as nib
import numpy as np
import torch
import torch.distributed as dist
from monai.transforms import Compose
from monai.utils import set_determinism

from .diff_model_setting import initialize_distributed, load_config, setup_logging
from .utils import define_instance

# Set the random seed for reproducibility
set_determinism(seed=0)


def create_transforms(dim: tuple = None) -> Compose:
    """

    Create a set of MONAI transforms for preprocessing.



    Args:

        dim (tuple, optional): New dimensions for resizing. Defaults to None.



    Returns:

        Compose: Composed MONAI transforms.

    """
    if dim:
        return Compose(
            [
                monai.transforms.LoadImaged(keys="image"),
                monai.transforms.EnsureChannelFirstd(keys="image"),
                monai.transforms.Orientationd(keys="image", axcodes="RAS"),
                monai.transforms.EnsureTyped(keys="image", dtype=torch.float32),
                monai.transforms.ScaleIntensityRanged(
                    keys="image", a_min=-1000, a_max=1000, b_min=0, b_max=1, clip=True
                ),
                monai.transforms.Resized(keys="image", spatial_size=dim, mode="trilinear"),
            ]
        )
    else:
        return Compose(
            [
                monai.transforms.LoadImaged(keys="image"),
                monai.transforms.EnsureChannelFirstd(keys="image"),
                monai.transforms.Orientationd(keys="image", axcodes="RAS"),
            ]
        )


def round_number(number: int, base_number: int = 128) -> int:
    """

    Round the number to the nearest multiple of the base number, with a minimum value of the base number.



    Args:

        number (int): Number to be rounded.

        base_number (int): Number to be common divisor.



    Returns:

        int: Rounded number.

    """
    new_number = max(round(float(number) / float(base_number)), 1.0) * float(base_number)
    return int(new_number)


def load_filenames(data_list_path: str) -> list:
    """

    Load filenames from the JSON data list.



    Args:

        data_list_path (str): Path to the JSON data list file.



    Returns:

        list: List of filenames.

    """
    with open(data_list_path, "r") as file:
        json_data = json.load(file)
    filenames_raw = json_data["training"]
    return [_item["image"] for _item in filenames_raw]


def process_file(

    filepath: str,

    args: argparse.Namespace,

    autoencoder: torch.nn.Module,

    device: torch.device,

    plain_transforms: Compose,

    new_transforms: Compose,

    logger: logging.Logger,

) -> None:
    """

    Process a single file to create training data.



    Args:

        filepath (str): Path to the file to be processed.

        args (argparse.Namespace): Configuration arguments.

        autoencoder (torch.nn.Module): Autoencoder model.

        device (torch.device): Device to process the file on.

        plain_transforms (Compose): Plain transforms.

        new_transforms (Compose): New transforms.

        logger (logging.Logger): Logger for logging information.

    """
    out_filename_base = filepath.replace(".gz", "").replace(".nii", "")
    out_filename_base = os.path.join(args.embedding_base_dir, out_filename_base)
    out_filename = out_filename_base + "_emb.nii.gz"

    if os.path.isfile(out_filename):
        return

    test_data = {"image": os.path.join(args.data_base_dir, filepath)}
    transformed_data = plain_transforms(test_data)
    nda = transformed_data["image"]

    dim = [int(nda.meta["dim"][_i]) for _i in range(1, 4)]
    spacing = [float(nda.meta["pixdim"][_i]) for _i in range(1, 4)]

    logger.info(f"old dim: {dim}, old spacing: {spacing}")

    new_data = new_transforms(test_data)
    nda_image = new_data["image"]

    new_affine = nda_image.meta["affine"].numpy()
    nda_image = nda_image.numpy().squeeze()

    logger.info(f"new dim: {nda_image.shape}, new affine: {new_affine}")

    try:
        out_path = Path(out_filename)
        out_path.parent.mkdir(parents=True, exist_ok=True)
        logger.info(f"out_filename: {out_filename}")

        with torch.amp.autocast("cuda"):
            pt_nda = torch.from_numpy(nda_image).float().to(device).unsqueeze(0).unsqueeze(0)
            z = autoencoder.encode_stage_2_inputs(pt_nda)
            logger.info(f"z: {z.size()}, {z.dtype}")

            out_nda = z.squeeze().cpu().detach().numpy().transpose(1, 2, 3, 0)
            out_img = nib.Nifti1Image(np.float32(out_nda), affine=new_affine)
            nib.save(out_img, out_filename)
    except Exception as e:
        logger.error(f"Error processing {filepath}: {e}")


@torch.inference_mode()
def diff_model_create_training_data(

    env_config_path: str, model_config_path: str, model_def_path: str, num_gpus: int

) -> None:
    """

    Create training data for the diffusion model.



    Args:

        env_config_path (str): Path to the environment configuration file.

        model_config_path (str): Path to the model configuration file.

        model_def_path (str): Path to the model definition file.

    """
    args = load_config(env_config_path, model_config_path, model_def_path)
    local_rank, world_size, device = initialize_distributed(num_gpus=num_gpus)
    logger = setup_logging("creating training data")
    logger.info(f"Using device {device}")

    autoencoder = define_instance(args, "autoencoder_def").to(device)
    try:
        checkpoint_autoencoder = torch.load(args.trained_autoencoder_path, weights_only=True)
        autoencoder.load_state_dict(checkpoint_autoencoder)
    except Exception:
        logger.error("The trained_autoencoder_path does not exist!")

    Path(args.embedding_base_dir).mkdir(parents=True, exist_ok=True)

    filenames_raw = load_filenames(args.json_data_list)
    logger.info(f"filenames_raw: {filenames_raw}")

    plain_transforms = create_transforms(dim=None)

    for _iter in range(len(filenames_raw)):
        if _iter % world_size != local_rank:
            continue

        filepath = filenames_raw[_iter]
        new_dim = tuple(
            round_number(
                int(plain_transforms({"image": os.path.join(args.data_base_dir, filepath)})["image"].meta["dim"][_i])
            )
            for _i in range(1, 4)
        )
        new_transforms = create_transforms(new_dim)

        process_file(filepath, args, autoencoder, device, plain_transforms, new_transforms, logger)

    if dist.is_initialized():
        dist.destroy_process_group()


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="Diffusion Model Training Data Creation")
    parser.add_argument(
        "--env_config",
        type=str,
        default="./configs/environment_maisi_diff_model_train.json",
        help="Path to environment configuration file",
    )
    parser.add_argument(
        "--model_config",
        type=str,
        default="./configs/config_maisi_diff_model_train.json",
        help="Path to model training/inference configuration",
    )
    parser.add_argument(
        "--model_def", type=str, default="./configs/config_maisi.json", help="Path to model definition file"
    )
    parser.add_argument("--num_gpus", type=int, default=1, help="Number of GPUs to use for distributed training")

    args = parser.parse_args()
    diff_model_create_training_data(args.env_config, args.model_config, args.model_def, args.num_gpus)