configs/train.yaml

# This defines the training script for the network
imports:
- $import os
- $import datetime
- $import torch
- $import scripts
- $import monai
- $import torch.distributed as dist
- $import operator

# Common elements to all training files
-
image: $monai.utils.CommonKeys.IMAGE
label: $monai.utils.CommonKeys.LABEL
pred: $monai.utils.CommonKeys.PRED

is_dist: '$dist.is_initialized()'
rank: '$dist.get_rank() if @is_dist else 0'
is_not_rank0: '$@rank > 0'
device: '$torch.device(f"cuda:{@rank}" if torch.cuda.is_available() else "cpu")'

network_def:
  _target_: monai.networks.nets.DiffusionModelUNet
  spatial_dims: 2
  in_channels: 1
  out_channels: 1
  channels: [64, 128, 128]
  attention_levels: [false, true, true]
  num_res_blocks: 1
  num_head_channels: 128

network: $@network_def.to(@device)
bundle_root: .
ckpt_path: $@bundle_root + '/models/model.pt'
use_amp: true
image_dim: 64
image_size: [1, '@image_dim', '@image_dim']
num_train_timesteps: 1000

base_transforms:
- _target_: LoadImaged
  keys: '@image'
  image_only: true
- _target_: EnsureChannelFirstd
  keys: '@image'
- _target_: ScaleIntensityRanged
  keys: '@image'
  a_min: 0.0
  a_max: 255.0
  b_min: 0.0
  b_max: 1.0
  clip: true

scheduler:
  _target_: monai.networks.schedulers.DDPMScheduler
  num_train_timesteps: '@num_train_timesteps'

inferer:
  _target_: monai.inferers.DiffusionInferer
  scheduler: '@scheduler'

# Training-specific

# choose a new directory for every run
output_dir: $datetime.datetime.now().strftime('./results/output_%y%m%d_%H%M%S')
dataset_dir: ./data

train_data:
  _target_ : MedNISTDataset
  root_dir: '@dataset_dir'
  section: training
  download: true
  progress: false
  seed: 0

val_data:
  _target_ : MedNISTDataset
  root_dir: '@dataset_dir'
  section: validation
  download: true
  progress: false
  seed: 0

train_datalist: '$[{"image": item["image"]} for item in @train_data.data if item["class_name"] == "Hand"]'
val_datalist: '$[{"image": item["image"]} for item in @val_data.data if item["class_name"] == "Hand"]'

batch_size: 8
num_substeps: 1
num_workers: 4
use_thread_workers: false

lr: 0.000025
rand_prob: 0.5
num_epochs: 75
val_interval: 5
save_interval: 5

train_transforms:
- _target_: RandAffined
  keys: '@image'
  rotate_range:
  - ['$-np.pi / 36', '$np.pi / 36']
  - ['$-np.pi / 36', '$np.pi / 36']
  translate_range:
  - [-1, 1]
  - [-1, 1]
  scale_range:
  - [-0.05, 0.05]
  - [-0.05, 0.05]
  spatial_size: [64, 64]
  padding_mode: "zeros"
  prob: '@rand_prob'

train_ds:
  _target_: Dataset
  data: $@train_datalist
  transform:
    _target_: Compose
    transforms: '$@base_transforms + @train_transforms'

train_loader:
  _target_: ThreadDataLoader
  dataset: '@train_ds'
  batch_size: '@batch_size'
  repeats: '@num_substeps'
  num_workers: '@num_workers'
  use_thread_workers: '@use_thread_workers'
  persistent_workers: '$@num_workers > 0'
  shuffle: true

val_ds:
  _target_: Dataset
  data: $@val_datalist
  transform:
    _target_: Compose
    transforms: '@base_transforms'

val_loader:
  _target_: DataLoader
  dataset: '@val_ds'
  batch_size: '@batch_size'
  num_workers: '@num_workers'
  persistent_workers: '$@num_workers > 0'
  shuffle: false

lossfn:
  _target_: torch.nn.MSELoss

optimizer:
  _target_: torch.optim.Adam
  params: $@network.parameters()
  lr: '@lr'

prepare_batch:
  _target_: monai.engines.DiffusionPrepareBatch
  num_train_timesteps: '@num_train_timesteps'

val_handlers:
- _target_: StatsHandler
  name: train_log
  output_transform: '$lambda x: None'
  _disabled_: '@is_not_rank0'

evaluator:
  _target_: SupervisedEvaluator
  device: '@device'
  val_data_loader: '@val_loader'
  network: '@network'
  amp: '@use_amp'
  inferer: '@inferer'
  prepare_batch: '@prepare_batch'
  key_val_metric:
    val_mean_abs_error:
      _target_: MeanAbsoluteError
      output_transform: $monai.handlers.from_engine([@pred, @label])
  metric_cmp_fn: '$operator.lt'
  val_handlers: '$list(filter(bool, @val_handlers))'

handlers:
- _target_: CheckpointLoader
  _disabled_: $not os.path.exists(@ckpt_path)
  load_path: '@ckpt_path'
  load_dict:
    model: '@network'
- _target_: ValidationHandler
  validator: '@evaluator'
  epoch_level: true
  interval: '@val_interval'
- _target_: CheckpointSaver
  save_dir: '@output_dir'
  save_dict:
    model: '@network'
  save_interval: '@save_interval'
  save_final: true
  epoch_level: true
  _disabled_: '@is_not_rank0'

trainer:
  _target_: SupervisedTrainer
  max_epochs: '@num_epochs'
  device: '@device'
  train_data_loader: '@train_loader'
  network: '@network'
  loss_function: '@lossfn'
  optimizer: '@optimizer'
  inferer: '@inferer'
  prepare_batch: '@prepare_batch'
  key_train_metric:
    train_acc:
      _target_: MeanSquaredError
      output_transform: $monai.handlers.from_engine([@pred, @label])
  metric_cmp_fn: '$operator.lt'
  train_handlers: '$list(filter(bool, @handlers))'
  amp: '@use_amp'

training:
- '$monai.utils.set_determinism(0)'
- '$@trainer.run()'