Spectogram classification

Description

This project includes all the code used during a private kaggle competition. This solution got 7th place on both public and private leaderboard.

Spectograms are split to 32 x 32 parts and the task is to classify which spectogram the part belongs to (there 3 parts).
Data is normalised to range [0, 1]
Achitecture consists of convolutional feature extractor and capsulenet head optimised according to this article.

For additional information / used data / trained models - send me a private message or an e-mail at kacp.kania@gmail.com.

Name		Name	Last commit message	Last commit date
Latest commit History 3 Commits
__pycache__		__pycache__
models		models
.gitignore		.gitignore
README.md		README.md
combine_results.py		combine_results.py
constants.py		constants.py
data.py		data.py
data_standard.py		data_standard.py
metrics.py		metrics.py
model.py		model.py
model_standard.py		model_standard.py
schedulers.py		schedulers.py
train.py		train.py
train_standard.py		train_standard.py
transformers.py		transformers.py
utils.py		utils.py