Curated segmentation papers from a field deep learning forgot

What it does

This is an annotated bibliography of time series segmentation research. It collects papers on segmentation, change-point detection, and anomaly detection—plus tangential work from information theory, computer vision, and motion capture—then sorts them by prominent research groups rather than by year or venue. The maintainer also flags recommended reads with a star emoji, which is about as opinionated as an awesome-list gets.

The interesting bit

The curator notes that segmentation has remained “lukewarm” compared to forecasting or classification, and that deep learning has not steamrolled the field; classic algorithms from a handful of research groups are still competitive. That honesty makes the list more useful than a generic dump of every arXiv preprint with “transformer” in the title.

Key highlights

• Organized by researcher and research group (e.g., David Hallac at Stanford), not chronologically.
• Explicitly covers adjacent topics: change-point detection, anomaly detection, and motion-capture segmentation.
• Breaks down methods by input type: univariate, multivariate, and tensor time series.
• Roughly 95% of the collected work is unsupervised, reflecting the lack of large public datasets with ground-truth segment labels.
• Includes surveys, benchmark datasets, evaluation frameworks, and code links where available.

Caveats

• The README is a work in progress and the author admits it carries “a degree of subjectivity.”
• Several listed papers have no linked implementation or dataset.
• The repository itself is a Markdown list, not a framework or tool.

Verdict

Worth bookmarking if you are doing literature review in temporal pattern discovery or need to justify why you chose a 2010s-era algorithm over a neural network. Skip it if you are looking for a ready-to-run Python package.