configs/inference.yaml

# This defines an inference script for generating a random image to a Pytorch file
imports:
- $import os
- $import datetime
- $import torch
- $import scripts
- $import monai
- $import torch.distributed as dist
- $import operator

# Common elements to all yaml 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'

# Inference-specific

batch_size: 1
num_workers: 0

noise: $torch.rand(1,1,@image_dim,@image_dim)  # create a random image every time this program is run

out_file: ""  # where to save the tensor to

# using a lambda this defines a simple sampling function used below
sample: '$lambda x: @inferer.sample(input_noise=x, diffusion_model=@network, scheduler=@scheduler)'

load_state: '$@network.load_state_dict(torch.load(@ckpt_path, weights_only = True))'  # command to load the saved model weights

save_trans:
  _target_: Compose
  transforms:
  - _target_: ScaleIntensity
    minv: 0.0
    maxv: 255.0
  - _target_: ToTensor
    track_meta: false
  - _target_: SaveImage
    output_ext: "jpg"
    resample: false
    output_dtype: '$torch.uint8'
    separate_folder: false
    output_postfix: '@out_file'

# program to load the model weights, run `sample`, and store results to `out_file`
testing:
- '@load_state'
- '$torch.save(@sample(@noise.to(@device)), @out_file)'

#alternative version which saves to a jpg file
testing_jpg:
- '@load_state'
- '$@save_trans(@sample(@noise.to(@device))[0])'