configs/inference_autoencoder.json

{
    "imports": [
        "$import torch",
        "$from datetime import datetime",
        "$from pathlib import Path"
    ],
    "bundle_root": ".",
    "model_dir": "$@bundle_root + '/models'",
    "dataset_dir": "/workspace/data/medical",
    "output_dir": "$@bundle_root + '/output'",
    "create_output_dir": "$Path(@output_dir).mkdir(exist_ok=True)",
    "device": "$torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')",
    "output_orig_postfix": "recon",
    "output_recon_postfix": "orig",
    "channel": 0,
    "spacing": [
        1.1,
        1.1,
        1.1
    ],
    "spatial_dims": 3,
    "image_channels": 1,
    "latent_channels": 8,
    "infer_patch_size": [
        144,
        176,
        112
    ],
    "autoencoder_def": {
        "_target_": "monai.networks.nets.autoencoderkl.AutoencoderKL",
        "spatial_dims": "@spatial_dims",
        "in_channels": "@image_channels",
        "out_channels": "@image_channels",
        "latent_channels": "@latent_channels",
        "channels": [
            64,
            128,
            256
        ],
        "num_res_blocks": 2,
        "norm_num_groups": 32,
        "norm_eps": 1e-06,
        "attention_levels": [
            false,
            false,
            false
        ],
        "with_encoder_nonlocal_attn": false,
        "with_decoder_nonlocal_attn": false,
        "include_fc": false
    },
    "load_autoencoder_path": "$@bundle_root + '/models/model_autoencoder.pt'",
    "load_autoencoder": "$@autoencoder_def.load_old_state_dict(torch.load(@load_autoencoder_path))",
    "autoencoder": "$@autoencoder_def.to(@device)",
    "preprocessing_transforms": [
        {
            "_target_": "LoadImaged",
            "keys": "image"
        },
        {
            "_target_": "EnsureChannelFirstd",
            "keys": "image"
        },
        {
            "_target_": "Lambdad",
            "keys": "image",
            "func": "$lambda x: x[@channel, :, :, :]"
        },
        {
            "_target_": "EnsureChannelFirstd",
            "keys": "image",
            "channel_dim": "no_channel"
        },
        {
            "_target_": "EnsureTyped",
            "keys": "image"
        },
        {
            "_target_": "Orientationd",
            "keys": "image",
            "axcodes": "RAS"
        },
        {
            "_target_": "Spacingd",
            "keys": "image",
            "pixdim": "@spacing",
            "mode": "bilinear"
        }
    ],
    "crop_transforms": [
        {
            "_target_": "CenterSpatialCropd",
            "keys": "image",
            "roi_size": "@infer_patch_size"
        }
    ],
    "final_transforms": [
        {
            "_target_": "ScaleIntensityRangePercentilesd",
            "keys": "image",
            "lower": 0,
            "upper": 99.5,
            "b_min": 0,
            "b_max": 1
        }
    ],
    "preprocessing": {
        "_target_": "Compose",
        "transforms": "$@preprocessing_transforms + @crop_transforms + @final_transforms"
    },
    "dataset": {
        "_target_": "monai.apps.DecathlonDataset",
        "root_dir": "@dataset_dir",
        "task": "Task01_BrainTumour",
        "section": "validation",
        "cache_rate": 0.0,
        "num_workers": 8,
        "download": false,
        "transform": "@preprocessing"
    },
    "dataloader": {
        "_target_": "DataLoader",
        "dataset": "@dataset",
        "batch_size": 1,
        "shuffle": true,
        "num_workers": 0
    },
    "saver_orig": {
        "_target_": "SaveImage",
        "_requires_": "@create_output_dir",
        "output_dir": "@output_dir",
        "output_postfix": "@output_orig_postfix",
        "resample": false,
        "padding_mode": "zeros"
    },
    "saver_recon": {
        "_target_": "SaveImage",
        "_requires_": "@create_output_dir",
        "output_dir": "@output_dir",
        "output_postfix": "@output_recon_postfix",
        "resample": false,
        "padding_mode": "zeros"
    },
    "input_img": "$monai.utils.first(@dataloader)['image'].to(@device)",
    "recon_img": "$@autoencoder(@input_img)[0][0]",
    "run": [
        "$@load_autoencoder",
        "$@saver_orig(@input_img[0][0])",
        "$@saver_recon(@recon_img)"
    ]
}