# 🥉 146th place – 9 models 1h inference - Segmentation through UNet

Short description of the solution below:
As I only used 1-3 days of time for modifying, training and tuning the solution stayed at 9 models 1h inference but could easily been more modified and with also more models for better score.
For the problem I used the base code from
[https://www.kaggle.com/code/tubotubo/cmi-submit](https://www.kaggle.com/code/tubotubo/cmi-submit)
[https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states](https://github.com/tubo213/kaggle-child-mind-institute-detect-sleep-states)
[https://www.kaggle.com/code/tubotubo/cmi-code](https://www.kaggle.com/code/tubotubo/cmi-code)
The original code had resnet34 as default and that was also the public pretrained model.  Based on that I trained several of different models with Spec2DCNN: Segmentation through UNet - resnet18, efficientnet-b0 and resnext101_32x4d.
Then I created a validation notebook below to get the validation score for some of the pretrained public and also the own trained models.
[https://www.kaggle.com/kirderf/eval-child-mind-institute-detect-sleep-states](https://www.kaggle.com/kirderf/eval-child-mind-institute-detect-sleep-states)
The final inference solution was a torch stacked mean ensemble of 9 models using 1h test time inference with the same duration used, code below.
[https://www.kaggle.com/kirderf/cmi-submit-9-models-1h-inference](https://www.kaggle.com/kirderf/cmi-submit-9-models-1h-inference)
.
With more time following had been tested: different architectures, more models,  augmentation, using different durations…
That’s it!
Happy Kaggling!