Ivadomed: Training Error: FILM layers value error

[vasudev-sharma]

I'm trying to follow the documentation, in particular, the tutorials to get started with training the models using ivadomed.

I'm following the tutorial of one class segmentation using 2D-unet. When I execute ivadomed -c config.json, I encounter following error log.

(ivadomed_env) p112473@bireli:~/p112473/ivadomed$ ivadomed -c config.json
Log directory already exists: spineGeneric
Using GPU number 1

Selected architecture: FiLMedUnet, with the following parameters:
        dropout_rate: 0.3
        bn_momentum: 0.9
        depth: 3
        is_2d: True
        folder_name: seg_sc_t1-t2-t2s-mt
        applied: True
        metadata: contrasts
        film_layers: [0, 1, 0, 0, 0, 0, 0, 0, 0, 0]
        in_channel: 1
        out_channel: 1

Selected transformations for the ['training'] dataset:
        Resample: {'wspace': 0.75, 'hspace': 0.75, 'dspace': 1}
        CenterCrop: {'size': [128, 128]}
        RandomAffine: {'degrees': 5, 'scale': [0.1, 0.1], 'translate': [0.03, 0.03], 'applied_to': ['im', 'gt']}
        ElasticTransform: {'alpha_range': [28.0, 30.0], 'sigma_range': [3.5, 4.5], 'p': 0.1, 'applied_to': ['im', 'gt']}
        NumpyToTensor: {}
        NormalizeInstance: {'applied_to': ['im']}

Selected transformations for the ['validation'] dataset:
        Resample: {'wspace': 0.75, 'hspace': 0.75, 'dspace': 1}
        CenterCrop: {'size': [128, 128]}
        NumpyToTensor: {}
        NormalizeInstance: {'applied_to': ['im']}
Loading dataset: 100%|███████████████████████████████████████████████████████████████████████████████████| 6/6 [00:00<00:00, 1545.15it/s]
Loaded 93 axial slices for the validation set.
Loading dataset: 100%|█████████████████████████████████████████████████████████████████████████████████| 18/18 [00:00<00:00, 1552.58it/s]
Loaded 291 axial slices for the training set.
Model directory already exists: spineGeneric/seg_sc_t1-t2-t2s-mt

Initialising model's weights from scratch.
Traceback (most recent call last):
  File "/home/GRAMES.POLYMTL.CA/p112473/miniconda/envs/ivadomed_env/bin/ivadomed", line 33, in <module>
    sys.exit(load_entry_point('ivadomed', 'console_scripts', 'ivadomed')())
  File "/home/GRAMES.POLYMTL.CA/p112473/p112473/ivadomed/ivadomed/main.py", line 311, in run_main
    resume_training=bool(args.resume_training))
  File "/home/GRAMES.POLYMTL.CA/p112473/p112473/ivadomed/ivadomed/main.py", line 216, in run_command
    debugging=context["debugging"])
  File "/home/GRAMES.POLYMTL.CA/p112473/p112473/ivadomed/ivadomed/training.py", line 101, in train
    model = model_class(**model_params)
  File "/home/GRAMES.POLYMTL.CA/p112473/p112473/ivadomed/ivadomed/models.py", line 522, in __init__
    "UNet depth. There should 2 * depth + 2 layers.".format(len(film_layers)))
ValueError: The number of FiLM layers 10 entered does not correspond to the UNet depth. There should 2 * depth + 2 layers.

NOTE: - The same data example and config file is used as mentioned in RTD, which is data_example_spinegenericand config.json respectively.

[andreanne-lemay]

Thank you for opening this issue!

Currently you are adding FiLM layers to your model. To avoid this error, you can simply disable this parameter by setting FiLMedUNet:applied to false:

    "FiLMedUnet": {
        "applied": false
}

By default it should be set to false. I fixed this in this PR .

[vasudev-sharma]

Thanks again @andreanne-lemay . I'll close this issue.

[jcohenadad]

issues are closed via PRs, not manually

[olix86]

It's too late for this one since the PR is already merged but good to know for the future

[charleygros]

It's too late for this one since the PR is already merged but good to know for the future

Agree. At least as far as I know.
So I will kindly close this issue. But if anyone knows how to proceed despite the PR already being merged: please feel to reopen it and proceed :-) (and I would be happy to learn this new tip!!)