Spaces:
Runtime error
Runtime error
File size: 5,579 Bytes
f85e212 |
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 |
from pathlib import Path
from datetime import datetime
import torch
from torch.utils.data import ConcatDataset
from pytorch_lightning.trainer import Trainer
from pytorch_lightning.callbacks import EarlyStopping, ModelCheckpoint
from medical_diffusion.data.datamodules import SimpleDataModule
from medical_diffusion.data.datasets import AIROGSDataset, MSIvsMSS_2_Dataset, CheXpert_2_Dataset
from medical_diffusion.models.embedders.latent_embedders import VQVAE, VQGAN, VAE, VAEGAN
import torch.multiprocessing
torch.multiprocessing.set_sharing_strategy('file_system')
if __name__ == "__main__":
# --------------- Settings --------------------
current_time = datetime.now().strftime("%Y_%m_%d_%H%M%S")
path_run_dir = Path.cwd() / 'runs' / str(current_time)
path_run_dir.mkdir(parents=True, exist_ok=True)
gpus = [0] if torch.cuda.is_available() else None
# ------------ Load Data ----------------
# ds_1 = AIROGSDataset( # 256x256
# crawler_ext='jpg',
# augment_horizontal_flip=True,
# augment_vertical_flip=True,
# # path_root='/home/gustav/Documents/datasets/AIROGS/dataset',
# path_root='/mnt/hdd/datasets/eye/AIROGS/data_256x256',
# )
# ds_2 = MSIvsMSS_2_Dataset( # 512x512
# # image_resize=256,
# crawler_ext='jpg',
# augment_horizontal_flip=True,
# augment_vertical_flip=True,
# # path_root='/home/gustav/Documents/datasets/Kather_2/train'
# path_root='/mnt/hdd/datasets/pathology/kather_msi_mss_2/train/'
# )
ds_3 = CheXpert_2_Dataset( # 256x256
# image_resize=128,
augment_horizontal_flip=False,
augment_vertical_flip=False,
# path_root = '/home/gustav/Documents/datasets/CheXpert/preprocessed_tianyu'
path_root = '/mnt/hdd/datasets/chest/CheXpert/ChecXpert-v10/preprocessed_tianyu'
)
# ds = ConcatDataset([ds_1, ds_2, ds_3])
dm = SimpleDataModule(
ds_train = ds_3,
batch_size=8,
# num_workers=0,
pin_memory=True
)
# ------------ Initialize Model ------------
model = VAE(
in_channels=3,
out_channels=3,
emb_channels=8,
spatial_dims=2,
hid_chs = [ 64, 128, 256, 512],
kernel_sizes=[ 3, 3, 3, 3],
strides = [ 1, 2, 2, 2],
deep_supervision=1,
use_attention= 'none',
loss = torch.nn.MSELoss,
# optimizer_kwargs={'lr':1e-6},
embedding_loss_weight=1e-6
)
# model.load_pretrained(Path.cwd()/'runs/2022_12_01_183752_patho_vae/last.ckpt', strict=True)
# model = VAEGAN(
# in_channels=3,
# out_channels=3,
# emb_channels=8,
# spatial_dims=2,
# hid_chs = [ 64, 128, 256, 512],
# deep_supervision=1,
# use_attention= 'none',
# start_gan_train_step=-1,
# embedding_loss_weight=1e-6
# )
# model.vqvae.load_pretrained(Path.cwd()/'runs/2022_11_25_082209_chest_vae/last.ckpt')
# model.load_pretrained(Path.cwd()/'runs/2022_11_25_232957_patho_vaegan/last.ckpt')
# model = VQVAE(
# in_channels=3,
# out_channels=3,
# emb_channels=4,
# num_embeddings = 8192,
# spatial_dims=2,
# hid_chs = [64, 128, 256, 512],
# embedding_loss_weight=1,
# beta=1,
# loss = torch.nn.L1Loss,
# deep_supervision=1,
# use_attention = 'none',
# )
# model = VQGAN(
# in_channels=3,
# out_channels=3,
# emb_channels=4,
# num_embeddings = 8192,
# spatial_dims=2,
# hid_chs = [64, 128, 256, 512],
# embedding_loss_weight=1,
# beta=1,
# start_gan_train_step=-1,
# pixel_loss = torch.nn.L1Loss,
# deep_supervision=1,
# use_attention='none',
# )
# model.vqvae.load_pretrained(Path.cwd()/'runs/2022_12_13_093727_patho_vqvae/last.ckpt')
# -------------- Training Initialization ---------------
to_monitor = "train/L1" # "val/loss"
min_max = "min"
save_and_sample_every = 50
early_stopping = EarlyStopping(
monitor=to_monitor,
min_delta=0.0, # minimum change in the monitored quantity to qualify as an improvement
patience=30, # number of checks with no improvement
mode=min_max
)
checkpointing = ModelCheckpoint(
dirpath=str(path_run_dir), # dirpath
monitor=to_monitor,
every_n_train_steps=save_and_sample_every,
save_last=True,
save_top_k=5,
mode=min_max,
)
trainer = Trainer(
accelerator='gpu',
devices=[0],
# precision=16,
# amp_backend='apex',
# amp_level='O2',
# gradient_clip_val=0.5,
default_root_dir=str(path_run_dir),
callbacks=[checkpointing],
# callbacks=[checkpointing, early_stopping],
enable_checkpointing=True,
check_val_every_n_epoch=1,
log_every_n_steps=save_and_sample_every,
auto_lr_find=False,
# limit_train_batches=1000,
limit_val_batches=0, # 0 = disable validation - Note: Early Stopping no longer available
min_epochs=100,
max_epochs=1001,
num_sanity_val_steps=2,
)
# ---------------- Execute Training ----------------
trainer.fit(model, datamodule=dm)
# ------------- Save path to best model -------------
model.save_best_checkpoint(trainer.logger.log_dir, checkpointing.best_model_path)
|