ValueError: No transformer cached unet

[neurocker]

I have created new environment and connect to neptune.ml, put some images to 'inference_directory' and run command to predict building borders:

python main.py predict-on-dir --pipeline_name unet_tta_scoring_model --chunk_size 1000 --dir_path C:\Users\admin\Documents\python\open-solution-mapping-challenge\data\inference_directory --prediction_path C:\Users\admin\Documents\python\open-solution-mapping-challenge\data\pred.json

and then error happened:

What I do wrong?

[neurocker]

Solution: download unet from here: https://ui.neptune.ml/neptune-ml/Mapping-Challenge/e/MC-1057/artifacts?file=unet Put file to open-solution-mapping-challenge\data\experiments\mapping_challenge_baseline\transformers

[jakubczakon]

Hi @bogdanov1ch! There is no trained unet model and so the code raises an exception. You need to either train first or use the pretrained model from this experiment.

Also, if you are using the unet_tta_scoring_model you need to train the second level model (lgbm) for scoring. I would suggest that you simply run the unet_tta which produces good results already and you don't have to train the second level.

I hope this helps.

[neurocker]

Thanks, @jakubczakon!