Why are the three weights stored?

[LimChaeYeon1003]

After completing the learning,
three weights are stored: ema, optimizer, and model.
Do we need all of this for testing? Where should I add it if necessary?

[Theodore-PKU]

Optimizer and model are stored because we may use them to train more steps.
ema is the average of models in the training process.
After training, model and ema are both able to be used for reconstruction (only use one of them).
In my experiment, using model is better.
Use kspace_test.py to run the model for reconstruction by adding --resume_checkpoint modelxxxxxx.pt. More detail can be seen in mri_recon_utils/kspace_script_util.py where all args needed are.

[LimChaeYeon1003]

Optimizer and model are stored because we may use them to train more steps. ema is the average of models in the training process. After training, model and ema are both able to be used for reconstruction (only use one of them). In my experiment, using model is better. Use kspace_test.py to run the model for reconstruction by adding --resume_checkpoint modelxxxxxx.pt. More detail can be seen in mri_recon_utils/kspace_script_util.py where all args needed are.

Good understanding. Thank you for answer