Skip to content

UPDATE.md

Latest commit

 

History

History
245 lines (166 loc) · 11.3 KB

UPDATE.md

File metadata and controls

245 lines (166 loc) · 11.3 KB
Raw

UPDATE.md — Refresh the Vidal Lab website

Branch note: on design-experiments the site uses a redesigned static-HTML layout driven by JS data. build.py is disabled. Edit pages and assets/js/*.js data files directly. The instructions below describe the legacy content/*.md + build.py flow on main.

Run this whenever the lab needs its homepage refreshed. The procedure is split into two independent tasks; do both unless the user says otherwise.

Conventions

  • All editable copy lives in content/*.md. Do not hand-edit index.html, people.html, tutorials.html, data-code.html, talks.html, etc. — they're generated.
  • After any change to content/, run python3 build.py to regenerate. Confirm "Done. Built 7 pages." (index, people, rene-vidal, teaching, talks, tutorials, data-code).
  • research.html is not rebuilt by build.py. To refresh publications, run python3 scripts/parse_bib.py. Its <nav> is also not regenerated — if you change navbar structure in build.py, hand-mirror the change in research.html.
  • The bibliography of record is vidal.bib. Dates referenced below come from arXiv IDs (YYMM.NNNNN) and the year = field.
  • Thumbnails live in two places. Keep them in sync:
    • assets/img/projects/<slug>.<ext> — used by the carousel slider.
    • assets/img/thumbnails/<slug>.<ext> — used by the highlights grid.
  • Never commit secrets, .claude/, or files >5 MB unless explicitly asked.

Site navigation

The top nav has Home / People / Research / Teaching plus a More dropdown containing Talks, Tutorials, and Data & Code. The dropdown is rendered by render_navbar() in build.py (and mirrored by hand in research.html). Tutorials and Data & Code are no longer on the home page — they live on tutorials.html and data-code.html. The More toggle gets .active automatically when the active page is talks, tutorials, or data-code.

Task 1 — Refresh project highlights

The home page has two parallel lists in content/index.md:

  1. project_carousel: — the rotating slider at the top.
  2. highlights: — the grid of cards lower down.

Both should stay in sync (same set of papers, same ordering newest-first).

Step 1.1 — Inventory what's currently shown

grep -nE 'title:|venue:' content/index.md | head -60

Note each highlight's title and venue. This is your "current set."

Step 1.2 — Find candidate new papers

grep -B1 -A12 'year\s*=\s*{2025\|year\s*=\s*{2026\|year\s*=\s*{2027' vidal.bib

(Use the current year and the previous year. Bump the years in this command as time passes.)

For each entry, record: BibTeX key, title, authors, venue (decode the @string like cvpr → "IEEE/CVF Conference on Computer Vision and Pattern Recognition" — for the website, use the short form like CVPR 2025), and the arXiv URL from Bib2Html_Dl_Pdf (convert /pdf/<id> → /abs/<id>).

A paper is a candidate if:

  • It's at a top-tier venue (CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, AISTATS, MICCAI, journals like TBME/PAMI), OR
  • The user explicitly mentions it.

Step 1.3 — Decide what to drop

Ask the user: "Currently highlighted: [list]. New candidates from the bib: [list]. Which to drop and which to add?" Offer a default of drop the oldest 1–3 entries; add the newest 1–3 candidates not already present.

Skip this prompt and use the default if the user said "auto" or "just refresh."

Step 1.4 — Download thumbnails for new entries

For each new paper, find a teaser image in this order of preference:

  1. Dedicated project page — search <author github.io username>/<paper slug> or look at the first author's personal site (e.g. hanchmin.github.io, peterljq.github.io, ryanchankh.github.io). Project pages usually have a static/images/teaser.png or assets/img/teaser.png.
  2. GitHub README — if the repo exists, raw images via https://raw.githubusercontent.com/<user>/<repo>/main/<path>.
  3. arXiv HTML version — https://arxiv.org/html/<id>v<latest>/x1.png is usually Figure 1.
  4. Author publication-preview — e.g. https://hanchmin.github.io/assets/img/publication_preview/<slug>.png.

Verify each download:

file <path>          # must report image data, not HTML
ls -la <path>        # must be > 5 KB and < 5 MB

If only an animation (>5 MB GIF) is available, downsize via ffmpeg or skip with a note.

Save into both directories with a slug filename:

cp /tmp/<paper>.png assets/img/thumbnails/<slug>.<ext>
cp /tmp/<paper>.png assets/img/projects/<slug>.<ext>

You may delegate the download work to a research subagent (general-purpose) — see the Voyaging/SECA precedent in git history.

Step 1.5 — Edit content/index.md

For each added paper, append/insert (in newest-first order) into both lists.

In project_carousel::

  - title: "<Paper title, short>"
    subtitle: "<Subtitle line, e.g. tagline or venue context>"
    authors: "<First, Second, ..., René Vidal>"
    venue: "<NeurIPS 2025>"
    description: "<One sentence, concrete about the contribution.>"
    image: assets/img/projects/<slug>.<ext>
    project_url: "<optional, omit if no project page>"
    paper_url: "https://arxiv.org/abs/<id>"

In highlights::

  - venue: "<NeurIPS 2025>"
    title: "<Full paper title>"
    authors: "<author list>"
    thumbnail: "assets/img/thumbnails/<slug>.<ext>"
    links:
      - text: "Project"          # optional
        url: "<project_url>"
      - text: "Paper"
        url: "https://arxiv.org/abs/<id>"

For each dropped paper, remove the corresponding YAML block from both lists.

Keep the two lists in the same order.

Step 1.6 — Rebuild and verify

python3 build.py

Open http://localhost:8000 (start python3 -m http.server 8000 in the background if not already running). Confirm:

  • Slider auto-advances every 6 s; prev/next/dots all work.
  • Each slide's image fills the left panel without cropping (object-fit: contain).
  • Highlights grid below shows the new cards with thumbnails not cropped.

Task 2 — Update news

The news list is news: in content/index.md. Each entry is date + text:

news:
  - date: "Feb 2026"
    text: "<short sentence>"

Step 2.1 — Identify new news items

Ask the user (or check provided sources) for:

  1. Paper acceptances — any new entries in vidal.bib since the last update whose venue isn't yet mentioned in news. Cross-reference: grep -E '"year' content/index.md won't help — instead diff vidal.bib against the most recent commit: 
    git log -1 --format=%H -- vidal.bib
git diff <that-hash>..HEAD -- vidal.bib | grep '^+@' | head
  2. Workshops & tutorials — new entries the user wants surfaced. Check content/talks.md and content/index.md's tutorials: list for additions.
  3. Speaker / invited-talk events for René Vidal — usually conveyed verbally. Confirm with the user; check content/talks.md for new entries.

If none of the above can be determined automatically, ask the user explicitly: "Anything new to add to News? (paper acceptances, workshops, talks)"

Step 2.2 — Style guide for news entries

  • date — month + year, e.g. "Feb 2026". Always quote.
  • text — one sentence, present tense, no trailing period when terse, with a period if the sentence is full. Match the surrounding entries' tone.
  • For multiple acceptances at the same venue, prefer one combined entry: "Three papers accepted to NeurIPS 2026!" rather than three separate ones.
  • Keep the list newest-first. Insert at the top.
  • Trim the bottom of the list if it grows past ~6 entries.

Step 2.3 — Apply and rebuild

Edit content/index.md, then:

python3 build.py

Verify the News section on the home page reflects the changes.

Task 3 — Update Tutorials & Workshops

Edit content/tutorials.md. Two parallel lists:

  • tutorials: — renders the Tutorials table on tutorials.html.
  • workshops: — renders the Workshops Organized table beneath it.

Each row uses the same shape:

  - year: "2025"
    title: "<short title>"
    venue: "<CVPR 2025>"
    url: "<page or empty string>"

Both tables use the .teaching-table style (Year / Title / Venue / Link). Keep entries newest-first. If url is empty the Link cell renders as "—".

Historical entries (2005–2017) were imported from vision.jhu.edu/activity.htm; preserve them when adding new ones.

Run python3 build.py and verify both tables render on tutorials.html.

Task 4 — Update Talks

Edit content/talks.md. Talks are grouped under sections:, each with a heading and a list of talks (title, venue, date). Add new invited talks / keynotes to the appropriate section, newest-first under "Recent Conference Talks." Run python3 build.py and confirm the talks.html output.

Task 5 — Update Data & Code

Edit content/data-code.md. The single data_code: list renders on data-code.html as <a> + description bullets. Run python3 build.py to regenerate.

Task 6 — Commit & push (only when the user asks)

Default behavior is never auto-commit. If the user says "commit and push":

git add content/ assets/img/projects/ assets/img/thumbnails/ index.html people.html
git commit -m "<short summary, e.g. 'refresh highlights and news'>"
git push origin main
git push upstream main

Both origin (ryanchankh/vidal-lab-website) and upstream (vidal-lab/vidal-lab.github.io) should track the same main.

If pushing to main is blocked by the auto-mode classifier, ask the user to run the push themselves with ! git push origin main && git push upstream main.

Troubleshooting

  • Carousel images zoomed/cropped — check .project-image { background-size: contain } and .highlight-thumb img { object-fit: contain } in assets/css/style.css.
  • Slider buttons/dots don't work — assets/js/main.js must count .project-slide (not just .carousel-slide/<img>); the count selector in the carousel init handles all three.
  • Image is HTML, not an image — re-download; some servers return a "captcha" or redirect HTML for hotlinks. Use a different source.
  • arXiv HTML 404 — try v1 instead of v2, or use arxiv.org/abs/<id> and screenshot/skip.
  • research.html shows wrong publications — run python3 scripts/parse_bib.py.
  • research.html navbar out of sync — research.html is not regenerated by build.py; update its <nav> and footer Links list by hand to match render_navbar() / FOOTER in build.py.

Recent structural changes

  • Talks tab in the navbar replaced with a More dropdown (Talks, Tutorials, Data & Code). Implemented in render_navbar(); mirrored in research.html. Styles in assets/css/style.css under /* ===== NAV DROPDOWN ===== */ plus mobile overrides; click-toggle wiring in assets/js/main.js.
  • tutorials.html and data-code.html are now standalone pages built from content/tutorials.md and content/data-code.md. The Tutorials and Data & Code sections were removed from content/index.md / index.html.
  • tutorials.html uses the Teaching-style table layout (.teaching-table) and includes a second Workshops Organized table. Historical entries (2005–2017) were imported from http://vision.jhu.edu/activity.htm.
  • Footer Links list extended to include Tutorials and Data & Code.