The MMAR (Medical Model ARchive) defines a standard structure for organizing all artifacts produced during the model development life cycle.
NVIDIA has example MMARs for liver, spleen, and others that can be found at NGC.
You can experiment with different configurations for the same MMAR. Create a new MMAR by cloning from an existing MMAR by using the cp -r OS command.
MMAR defines the standard structure for storing artifacts (files) needed and produced by the model development workflow (training, validation, inference, etc.). The MMAR includes all the information about the model including configurations and scripts to provide a work space to perform all model development tasks:
ROOT
config
config_train.json
config_finetune.json
config_inference.json
config_validation.json
config_validation_ckpt.json
environment.json
commands
set_env.sh
train.sh train with single GPU
train_multi_gpu.sh train with 2 GPUs
finetune.sh transfer learning with CKPT
infer.sh inference with TS model
validate.sh valiidate with TS model
validate_ckpt.sh validate with CKPT
validate_multi_gpu.sh validate with TS model on 2 GPUs
validate_multi_gpu_ckpt.sh validate with CKPT on 2 GPUs
export.sh export CKPT to TS model
resources
log.config
...
docs
license.txt
Readme.md
...
models
model.pt
model.ts
final_model.pt
eval
all evaluation outputs: segmentation / classification results
metrics reports, etc.
“CKPT” means regular PyTorch “.pt” weights, “TS” means PyTorch TorchScript model
Clara workflows are made of different types of components. For each type, there are usually multiple choices. To put together a workflow, you specify and configure the components to be used.
Clara offers training and validation workflows. Workflow configurations are defined in JSON files: for example config_train.json for training workflows and config_validation.json for validation workflows.
Training configuration
The training config file config_train.json defines the configuration of the training workflow. The config contains three sections: global variables, train, and validate.
You can define global variables in the configuration JSON file. These variables can be overwritten through environment.json and then the shell command or command line at the highest priority.
For an example to reference, see the config_train.json of the clara_pt_spleen_ct_segmentation MMAR.
The “train” section defines the components for the training process, including “loss”, “optimizer”, “lr_scheduler”, “model”, “pre_transforms”, “dataset”, “dataloader”, “inferer”, “handlers”, “post_transforms”, “metrics” and “trainer”. Each component is constructed by providing the component’s class “name” and the init arguments “args”.
Similarly, the “validate” section defines the components for validation process, including “pre_transforms”, “dataset”, “dataloader”, “inferer”, “handlers”, “post_transforms”, “metrics”, and “evaluator”. Each component is constructed the same way by providing the component class “name” and the corresponding init arguments “args”.
From Clara v4.0 on compared to v3.X, almost all components are now from the open source project MONAI. All the MONAI API docstrings are available at: https://docs.monai.io/en/latest/.
The training and validation workflows are based on MONAI workflows which are developed based on ignite.
The provided commands perform model development work based on the configurations in the config folder. The only command you may need to change is set_env.sh, where you can set the PYTHONPATH to the proper value.
You don’t need to change any other commands for default behavior, but you can and should study them to understand how they are defined.
Please see the commands in the example MMARs for the most up to date features and settings that can be used.
train.sh
This command is used to do basic single-gpu training from scratch. When finished, you should see the following files in the “models” folder:
model.pt- the best model obtainedfinal_model.pt- the final model when the training is done. It is usually NOT the best model.Event files - these are tensorboard events that you can view with tensorboard.
Example: train.sh
1 #!/usr/bin/env bash
2 my_dir="$(dirname "$0")"
3 . $my_dir/set_env.sh
4 echo "MMAR_ROOT set to $MMAR_ROOT"
5
6 CONFIG_FILE=config/config_train.json
7 ENVIRONMENT_FILE=config/environment.json
8 python3 -u -m medl.apps.train \
9 -m $MMAR_ROOT \
10 -c $CONFIG_FILE \
11 -e $ENVIRONMENT_FILE \
12 --set \
13 DATASET_JSON=$MMAR_ROOT/config/dataset_0.json \
14 epochs=1260 \
15 multi_gpu=false \
16 ${additional_options}
Line numbers are not part of the command.
Explanation: train.sh
Line 1: this is a bash script
Lines 2 to 3: resolve and set the absolute directory path for MMAR_ROOT
Line 6: set the training config file
Line 7: set the environment file that defines commonly used variables
such as DATA_ROOT.
Lines 8 to 17: invokes the training program.
Lines 9 to 11: set the arguments required by the training program
Line 12: the --set directive allows certain training parameters to be overwritten
Line 13: set the DATASET_JSON to use the dataset_0.json in the “config” folder of
the MMAR. This overwrites the DATASET_JSON defined in the environment.json file.
Lines 14 to 16: overwrite the training variables as defined in config_train.json
train_multi_gpu.sh
Example: train_multi_gpu.sh
1 #!/usr/bin/env bash
2
3 my_dir="$(dirname "$0")"
4 . $my_dir/set_env.sh
5
6 echo "MMAR_ROOT set to $MMAR_ROOT"
7 additional_options="$*"
8
9 CONFIG_FILE=config/config_train.json
10 ENVIRONMENT_FILE=config/environment.json
11
12 python -m torch.distributed.launch --nproc_per_node=2 --nnodes=1 --node_rank=0 \
13 --master_addr="localhost" --master_port=1234 \
14 -m medl.apps.train \
15 -m $MMAR_ROOT \
16 -c $CONFIG_FILE \
17 -e $ENVIRONMENT_FILE \
18 --write_train_stats \
19 --set \
20 print_conf=True \
21 epochs=1260 \
22 multi_gpu=True \
23 ${additional_options}
Explanation: train_multi_gpu.sh
This command uses torch.distributed to launch multiple gpu training.
finetune.sh
This command is used to continue training from previous model. Before running this command, you must have a previous model in the models folder. The output of this command is the same as train.sh.
Example: finetune.sh
1 #!/usr/bin/env bash
2 my_dir="$(dirname "$0")"
3 . $my_dir/set_env.sh
4 echo "MMAR_ROOT set to $MMAR_ROOT"
5 additional_options="$*"
6
7 CONFIG_FILE=config/config_finetune.json
8 ENVIRONMENT_FILE=config/environment.json
9
11 python3 -u -m medl.apps.train \
12 -m $MMAR_ROOT \
13 -c $CONFIG_FILE \
14 -e $ENVIRONMENT_FILE \
15 --write_train_stats \
16 --set \
17 print_conf=True \
18 epochs=1260 \
19 multi_gpu=False \
20 ${additional_options}
export.sh
Export a trained model checkpoint to frozen graphs. Two frozen graphs will be generated into the models folder.
Training must have been done before you run this command.
model.fzn.pb - the regular frozen graph
model.trt.pb - TRT-optimized frozen graph
Example: export.sh
1 #!/usr/bin/env bash
2 my_dir="$(dirname "$(readlink -f "$0")")"
3 . $my_dir/set_env.sh
4
5 CONFIG_FILE=config/config_train.json
6 ENVIRONMENT_FILE=config/environment.json
7 export CKPT_DIR=$MMAR_ROOT/models
8
9 INPUT_CKPT="${MMAR_ROOT}/models/model.pt"
10 OUTPUT_CKPT="${MMAR_ROOT}/models/model.ts"
6 python3 -u -m medl.apps.export \
7 -m $MMAR_ROOT \
8 -c $CONFIG_FILE \
9 -e $ENVIRONMENT_FILE \
10 --model_path "${INPUT_CKPT}" \
11 --output_path "${OUTPUT_CKPT}" \
12 --input_shape "[1, 1, 160, 160, 160]"
Explanation: export.sh
Lines 6 to 12: invoke the export program.
infer.sh
Perform inference against the model, based on the configuration of config_validation.json in the config folder. Inference output is saved in the eval folder.
Example: infer.sh
1 #!/usr/bin/env bash
2 my_dir="$(dirname "$0")"
3 . $my_dir/set_env.sh
4 echo "MMAR_ROOT set to $MMAR_ROOT"
5
6 CONFIG_FILE=config/config_validation.json
7 ENVIRONMENT_FILE=config/environment.json
8 python3 -u -m medl.apps.evaluate \
9 -m $MMAR_ROOT \
10 -c $CONFIG_FILE \
11 -e $ENVIRONMENT_FILE \
12 --set \
13 DATASET_JSON=$MMAR_ROOT/config/dataset_0.json \
14 output_infer_result=true \
15 do_validation=false
Explanation: infer.sh
Line 1: this is a bash script
Lines 2 to 3: resolve and set the absolute directory path for MMAR_ROOT
Line 6: set the validation config file
Line 7: set the environment file that defines commonly used variables such as DATA_ROOT.
Lines 8 to 15: invokes the evaluate program.
Lines 9 to 11: set the arguments required by the program
Line 12: the --set directive allows certain parameters to be overwritten
Line 13: set the DATASET_JSON to use the dataset_0.json in the “config” folder of the MMAR. This overwrites the DATASET_JSON defined in the environment.json file.
validate.sh
Perform validation against the model, based on the configuration of config_validation.json in the config folder. Validation output is saved in the evalfolder.
Example: validate.sh
1 #!/usr/bin/env bash
2 my_dir="$(dirname "$0")"
3 . $my_dir/set_env.sh
4 echo "MMAR_ROOT set to $MMAR_ROOT"
5
6 CONFIG_FILE=config/config_validation.json
7 ENVIRONMENT_FILE=config/environment.json
8 python3 -u -m medl.apps.evaluate \
9 -m $MMAR_ROOT \
10 -c $CONFIG_FILE \
11 -e $ENVIRONMENT_FILE
Explanation: validate.sh
This command is very similar to infer.sh. The only differences are:
infer.sh and validate.sh use the same evaluate program but different configurations.
The JSON files in the config folder define configurations of workflow tasks (training, inference, and validation).
config_train.json
This file defines components that make up the training workflow. It is used by all four training commands (single-gpu training and finetuning, multi-gpu training and finetuning). See Training configuration for details.
config_validation.json
This file defines configuration that is used for both validate.sh and infer.sh. The only difference between the two commands are the options of do_validation and output_infer_result. See Validation configuration for details on the configuration file for validation.
environment.json
This file defines the common parameters for all model work. The most important are DATA_ROOT and DATASET_JSON.
DATA_ROOT specifies the directory that contains the training data.
DATASET_JSON specifies the config file that contains the default training data split (usually dataset_0.json).
Since MMAR does not contain training data, you must ensure that these two parameters are set to the right value. Do not change any other parameters.
Example: environment.json
{
"DATA_ROOT": "/workspace/data/Task09_Spleen_nii",
"DATASET_JSON": "/workspace/data/Task09_Spleen_nii/dataset_0.json",
"PROCESSING_TASK": "segmentation",
"MMAR_EVAL_OUTPUT_PATH": "eval",
"MMAR_CKPT_DIR": "models",
"MMAR_CKPT": "models/model.pt"
"MMAR_TORCHSCRIPT": "models/models/ts"
}
Variable |
Description |
|---|---|
DATA_ROOT |
The location of training data |
DATASET_JSON |
The data split config file |
PROCESSING_TASK |
The task type of the training: segmentation or classification |
MMAR_EVAL_OUTPUT_PATH |
Directory for saving evaluation (validate or infer) results. Always the “eval” folder in the MMAR |
MMAR_CKPT_DIR |
Directory for saving training results. Always the “models” folder in the MMAR |
Data split config file
This is a JSON file that defines the data split used for training and validation. For classification model, this file is usually named “plco.json”; for other models, it is usually named “dataset_0.json”.
The following is dataset_0.json of the model segmentation_ct_spleen:
{
"description": "Spleen Segmentation",
"labels": {
"0": "background",
"1": "spleen"
},
"licence": "CC-BY-SA 4.0",
"modality": {
"0": "CT"
},
"name": "Spleen",
"numTest": 20,
"numTraining": 41,
"reference": "Memorial Sloan Kettering Cancer Center",
"release": "1.0 06/08/2018",
"tensorImageSize": "3D",
"training": [
{
"image": "imagesTr/spleen_29.nii.gz",
"label": "labelsTr/spleen_29.nii.gz"
},
… <<more data here>>…..
{
"image": "imagesTr/spleen_49.nii.gz",
"label": "labelsTr/spleen_49.nii.gz"
}
],
"validation": [
{
"image": "imagesTr/spleen_19.nii.gz",
"label": "labelsTr/spleen_19.nii.gz"
},
… <<more data here>>…..
{
"image": "imagesTr/spleen_9.nii.gz",
"label": "labelsTr/spleen_9.nii.gz"
}
]
}
There is a lot of information in this file, but the only sections needed by the training and validation programs are the “training” and “validation” sections, which define sample/label pairs of data items for training and validation respectively.
In the models folder, model.pt is the best model resulting from training. final_model.pt is created when the training is finished normally. final_model is a snapshot of the model at the last moment. It is usually not the best model that can be obtained. Both model.pt and final_model.pt can be used for further training or fine-tuning. Here are two typical use cases:
Continued training: Use the
final_model.ptas the starting point for fine-tuning if you think the model has not converged due to improper configuration with the number of epochs not set high enough.Transfer learning: Use the
model.ptas the starting point for fine-tuning on a different dataset, which may be your own dataset, to obtain the model that is best for your data. This is also called adaptation.
A MMAR is a self-contained workspace for model development work. If you want to experiment with different configurations for the same MMAR, you should create a new MMAR by cloning from an existing MMAR.