skself

Developing Self-supervised Systems in an industrial Setting

The goal of this project is to explore methods to reduce the amount of effort required for industries to implement machine learning methods either by reducing the effort required to curate and annotate their datasets or by exploring out-of-the-box solutions like multimodal large language models or pretrained embeddings.

Features

• Lazy Labels:

  skself.partial_annotations.lazy_model.LazySegmentationModel
  

• GPT Anomaly Detection:

  skself.partial_annotations MLLMANO.ipynb
  

• Embedding Training:

  skself.embedding_training.embedding_benchmark.Baseline
  

Publications

Method	Paper	Link
Partial Annotations	Lazy Labels for Segmentation	Link 1
Multimodal Large Language Models	Low-shot Visual Anomaly Detection with Multimodal Large Language Models	Link 2

Install

Create a new python=3.9 env and install skself from pip

pip install git+https://github.com/thetoby9944/skself.git

Examples

To directly jump into the code look at the sample notebook

Cite

If this project helped you during your work: Until a publication is available, please cite as

Tobias Schiele et al. (2023). skself https://github.com/thetoby9944/skself.

@misc{Schiele2019,
    author = {Tobias Schiele, Daria Kern, Prof. Dr. Ulrich Klauck},
    title = {Skself},
    year = {2022},
    publisher = {GitHub},
    journal = {GitHub repository},
    howpublished = {\url{https://github.com/thetoby9944/skself}},
}

This work is funded by

until July 2024 and will receive continued personal updates and maintenance over the course of my PhD until January 2025.