[Auto3D3Seg] Saved model.pt/final_model.pt have missing keys when loaded in comparison with SegRestNetDS (mentioned in hyperparameters)

[spes3]

I am trying to convert my trained model to torchscript model for deployment, I keep getting that there are missed keys.

My code till now:

from monai.networks.nets import SegResNetDS


  model = SegResNetDS(
    init_filters=32,
    blocks_down=[1, 2, 2, 4, 4],
    norm='BATCH',
    in_channels=1,
    out_channels=2,
    dsdepth=4 )

model.load_state_dict(torch.load(model_path/model_final.pt)['state_dict'])

model.eval() 

The output/Error:

RuntimeError: Error(s) in loading state_dict for SegResNetDS:
        Missing key(s) in state_dict: "encoder.layers.0.blocks.0.norm1.running_mean", "encoder.layers.0.blocks.0.norm1.running_var", "encoder.layers.0.blocks.0.norm1.num_batches_tracked", "encoder.layers.0.blocks.0.norm2.running_mean", "encoder.layers.0.blocks.0.norm2.running_var", "encoder.layers.0.blocks.0.norm2.num_batches_tracked", "encoder.layers.1.blocks.0.norm1.running_mean", "encoder.layers.1.blocks.0.norm1.running_var", "encoder.layers.1.blocks.0.norm1.num_batches_tracked", "encoder.layers.1.blocks.0.norm2.running_mean", "encoder.layers.1.blocks.0.norm2.running_var", "encoder.layers.1.blocks.0.norm2.num_batches_tracked", "encoder.layers.1.blocks.1.norm1.running_mean", "encoder.layers.1.blocks.1.norm1.running_var", "encoder.layers.1.blocks.1.norm1.num_batches_tracked", "encoder.layers.1.blocks.1.norm2.running_mean", 
"encoder.layers.1.blocks.1.norm2.running_var", "encoder.layers.1.blocks.1.norm2.num_batches_tracked", "encoder.layers.2.blocks.0.norm1.running_mean", "encoder.layers.2.blocks.0.norm1.running_var", "encoder.layers.2.blocks.0.norm1.num_batches_tracked", "encoder.layers.2.blocks.0.norm2.running_mean", "encoder.layers.2.blocks.0.norm2.running_var", "encoder.layers.2.blocks.0.norm2.num_batches_tracked", "encoder.layers.2.blocks.1.norm1.running_mean", "encoder.layers.2.blocks.1.norm1.running_var", "encoder.layers.2.blocks.1.norm1.num_batches_tracked", "encoder.layers.2.blocks.1.norm2.running_mean", "encoder.layers.2.blocks.1.norm2.running_var", "encoder.layers.2.blocks.1.norm2.num_batches_tracked", "encoder.layers.3.blocks.0.norm1.running_mean", "encoder.layers.3.blocks.0.norm1.running_var", "encoder.layers.3.blocks.0.norm1.num_batches_tracked", "encoder.layers.3.blocks.0.norm2.running_mean", "encoder.layers.3.blocks.0.norm2.running_var", "encoder.layers.3.blocks.0.norm2.num_batches_tracked", "encoder.layers.3.blocks.1.norm1.running_mean", "encoder.layers.3.blocks.1.norm1.running_var", "encoder.layers.3.blocks.1.norm1.num_batches_tracked", "encoder.layers.3.blocks.1.norm2.running_mean", 
"encoder.layers.3.blocks.1.norm2.running_var", "encoder.layers.3.blocks.1.norm2.num_batches_tracked", "encoder.layers.3.blocks.2.norm1.running_mean", "encoder.layers.3.blocks.2.norm1.running_var", "encoder.layers.3.blocks.2.norm1.num_batches_tracked", "encoder.layers.3.blocks.2.norm2.running_mean", "encoder.layers.3.blocks.2.norm2.running_var", "encoder.layers.3.blocks.2.norm2.num_batches_tracked", "encoder.layers.3.blocks.3.norm1.running_mean", "encoder.layers.3.blocks.3.norm1.running_var", "encoder.layers.3.blocks.3.norm1.num_batches_tracked", "encoder.layers.3.blocks.3.norm2.running_mean", "encoder.layers.3.blocks.3.norm2.running_var", "encoder.layers.3.blocks.3.norm2.num_batches_tracked", "encoder.layers.4.blocks.0.norm1.running_mean", "encoder.layers.4.blocks.0.norm1.running_var", "encoder.layers.4.blocks.0.norm1.num_batches_tracked", "encoder.layers.4.blocks.0.norm2.running_mean", "encoder.layers.4.blocks.0.norm2.running_var", "encoder.layers.4.blocks.0.norm2.num_batches_tracked", "encoder.layers.4.blocks.1.norm1.running_mean", "encoder.layers.4.blocks.1.norm1.running_var", "encoder.layers.4.blocks.1.norm1.num_batches_tracked", "encoder.layers.4.blocks.1.norm2.running_mean", "encoder.layers.4.blocks.1.norm2.running_var", "encoder.layers.4.blocks.1.norm2.num_batches_tracked", "encoder.layers.4.blocks.2.norm1.running_mean", "encoder.layers.4.blocks.2.norm1.running_var", "encoder.layers.4.blocks.2.norm1.num_batches_tracked", "encoder.layers.4.blocks.2.norm2.running_mean", "encoder.layers.4.blocks.2.norm2.running_var", "encoder.layers.4.blocks.2.norm2.num_batches_tracked", 
"encoder.layers.4.blocks.3.norm1.running_mean", "encoder.layers.4.blocks.3.norm1.running_var", "encoder.layers.4.blocks.3.norm1.num_batches_tracked", "encoder.layers.4.blocks.3.norm2.running_mean", "encoder.layers.4.blocks.3.norm2.running_var", "encoder.layers.4.blocks.3.norm2.num_batches_tracked", 
"up_layers.0.blocks.0.norm1.running_mean", "up_layers.0.blocks.0.norm1.running_var", "up_layers.0.blocks.0.norm1.num_batches_tracked", "up_layers.0.blocks.0.norm2.running_mean", "up_layers.0.blocks.0.norm2.running_var", "up_layers.0.blocks.0.norm2.num_batches_tracked", "up_layers.1.blocks.0.norm1.running_mean", "up_layers.1.blocks.0.norm1.running_var", "up_layers.1.blocks.0.norm1.num_batches_tracked", "up_layers.1.blocks.0.norm2.running_mean", "up_layers.1.blocks.0.norm2.running_var", "up_layers.1.blocks.0.norm2.num_batches_tracked", "up_layers.2.blocks.0.norm1.running_mean", "up_layers.2.blocks.0.norm1.running_var", "up_layers.2.blocks.0.norm1.num_batches_tracked", "up_layers.2.blocks.0.norm2.running_mean", "up_layers.2.blocks.0.norm2.running_var", "up_layers.2.blocks.0.norm2.num_batches_tracked", "up_layers.3.blocks.0.norm1.running_mean", "up_layers.3.blocks.0.norm1.running_var", "up_layers.3.blocks.0.norm1.num_batches_tracked", "up_layers.3.blocks.0.norm2.running_mean", "up_layers.3.blocks.0.norm2.running_var", "up_layers.3.blocks.0.norm2.num_batches_tracked".

Is there anyting I'm missing? In the hyper_parameters.yaml the only difference is the norm for SegResNetDS. I tried this with simulated data and my images and with autorunner and with the one input training/inference pipeline and all the models somehow don't match? Thank you

[spes3]

i shouldve used norm = "instance"
this solved everything