Results page for TransfHAR: Self-Supervised Wrist Representations for On-Demand Activity Recognition (UIST '26). One view for everyone: headline results, an activity explorer (pick any of 127 activities to see how well it is recognized and what it is confused with) and a leaderboard. The selected activity is kept in the URL so results can be shared as links. Design source: Claude Design project "TransfHAR Results".
npm install
npm run dev # http://localhost:5173
npm run build # static site in dist/Set BASE when the site is served from a subpath, e.g. BASE=/TransfHAR/ npm run build.
Everything is rendered at runtime from public/data/:
paper.json: curated from the paper (tables, stated numbers, taxonomy, user-study design, activity bank). Values the paper doesn't state numerically are left out rather than read off plots.confusions.json: the Figure 14 (SAMoSA, UTD-MHAD, PrISM latte) and Figure 15 (P1–P10) confusion matrices, extracted from the PDF text bynpm run data(scripts/extract_confusions.py, needs poppler'spdftotext).
res/TransfHAR.pdf is the source paper. The research code (SPICExLAB/IMU-SSL, private) can be checked out at res/IMU-SSL; it is git-ignored and not needed to build the site.
.github/workflows/pages.yml builds and deploys to GitHub Pages on every push to main. Enable Pages with
"GitHub Actions" as the source. For a custom domain, set the repository variable PAGES_BASE to /.