Add static site and GitHub Pages deployment

The Dash app needs a Python server, which rules out free static hosting.
Add a static front-end that renders the same map client-side:

- build_site.py merges papers + layout into a single column-oriented
  site/data.json (7 MB, generated, gitignored)
- site/index.html draws it with plotly.js from CDN, plus keyword search
  over titles and abstracts with a shareable ?q= URL param
- .github/workflows/pages.yml rebuilds and publishes site/ on push to main

app.py is kept for local use. README rewritten with fork attribution.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: coolteemf

.github/workflows/pages.yml

@@ -0,0 +1,32 @@
+name: Deploy site to GitHub Pages
+
+on:
+  push:
+    branches: [main]
+  workflow_dispatch:
+
+permissions:
+  contents: read
+  pages: write
+  id-token: write
+
+concurrency:
+  group: pages
+  cancel-in-progress: true
+
+jobs:
+  deploy:
+    runs-on: ubuntu-latest
+    environment:
+      name: github-pages
+      url: ${{ steps.deployment.outputs.page_url }}
+    steps:
+      - uses: actions/checkout@v4
+      - name: Build site payload
+        run: python build_site.py
+      - uses: actions/configure-pages@v5
+      - uses: actions/upload-pages-artifact@v3
+        with:
+          path: site
+      - id: deployment
+        uses: actions/deploy-pages@v4

.gitignore

@@ -1 +1,2 @@
 .venv/
+site/data.json

README.md

@@ -1,47 +1,79 @@
-# cvpr-explorer
-Visualize CVPR papers as embeddings on [cvprexplorer.com](http://cvprexplorer.com)
+# CVPR 2026 Explorer
 
-# Problem Statement
-As the field of AI and machine learning has exploded in recent years, it has become increasingly difficult to keep up to date with latest research. Conference websites frequently list >2000 papers in one giant list out of which it is not easy to find papers one is interested in.
+A 2D semantic map of all 4068 CVPR 2026 papers, built from
+[SPECTER2](https://huggingface.co/allenai/specter2) embeddings of their abstracts.
+Nearby points are semantically similar papers, so a cluster is a topic.
 
-# This Project
-The current project is a visual aid for browsing CVPR papers. I am starting with CVPR because its a popular conference and has extensive public records. I scraped the website and found the paper titles and abstracts, then used OpenAI embedding API to find vector embeddings of the abstracts. Finally, I visualized the embeddings in 2D with t-SNE plot. Each point in the plot is a paper.
+**Live site: https://flecomet.github.io/cvpr-explorer/**
 
-The results are amazing. The embeddings reveal clusters of popular topics like transformers, NeRF, diffusion models etc in recent years. So, if you are interested in only diffusion models, for example, you do not have to wade through hundreds of irrelevant papers and only need to look at points (papers) in the corresponding cluster.
+> This is a fork of [dataplayer12/cvpr-explorer](https://github.com/dataplayer12/cvpr-explorer)
+> ([cvprexplorer.com](http://cvprexplorer.com)), rewritten for CVPR 2026: OpenAI embeddings
+> replaced with the open SPECTER2 scientific-document model, automatic cluster labels, and a
+> static front-end that needs no server. Original idea and design credit to
+> [@dataplayer12](https://github.com/dataplayer12). Same license as upstream — see [LICENSE](LICENSE).
 
-I hope this tool will be useful for stuents, researchers and ML engineers. I invite the community to biuld on this project and add more conferences. I hope that some day this will become the standard way of finding new papers and that unorganized long lists of >2000 papers will be a thing of the past.
+## Why
 
-# Demo
-You can view the app at [cvprexplorer.com](http://cvprexplorer.com)
+Conference sites list 4000+ papers as one flat list. Finding the ones you care about means
+scrolling past thousands you don't. Here you zoom into the region of the map you're interested
+in — diffusion, 3D reconstruction, medical imaging — and read only those.
 
-# Results
-![embeddings](https://github.com/dataplayer12/cvpr-explorer/assets/11517109/d8913d58-26f2-43a2-b0f8-de5c526f0ab6)
+## How it works
 
+The pipeline is offline; the site is static.
 
-## Deploy using docker
-```Shell
-git clone http://github.com/dataplayer12/cvpr-explorer.git
-cd cvpr-explorer/
-docker build -t cvpr-explorer .
-docker run --network=host cvpr-explorer
+| Step | Script | Output |
+|------|--------|--------|
+| Scrape titles, authors, abstracts, PDF links | `scrape.py` | `data/cvpr_2026_papers.json` |
+| Embed abstracts with SPECTER2 | `embed.py` | `data/cvpr_2026_specter2.npy` |
+| UMAP to 2D + HDBSCAN clusters + c-TF-IDF labels | `layout.py` | `data/cvpr_2026_layout.json` |
+| Merge into the site payload | `build_site.py` | `site/data.json` |
+
+`site/index.html` renders that payload client-side with plotly.js. No backend, no API keys at
+serve time.
+
+## Run the pipeline
+
+```shell
+pip install -r requirements-pipeline.txt
+python scrape.py     # writes data/cvpr_2026_papers.json
+python embed.py      # writes data/cvpr_2026_specter2.npy (GPU recommended)
+python layout.py     # writes data/cvpr_2026_layout.json
+```
+
+## Serve locally
+
+Static site (what the live URL serves):
+
+```shell
+python build_site.py
+python -m http.server -d site 8000   # http://localhost:8000
 ```
 
-## Deploy with docker compose
-```Shell
+Or the original Dash app:
+
+```shell
+pip install -r requirements.txt
+python app.py                        # http://localhost:8050
+```
+
+Or with Docker:
+
+```shell
 docker compose build
 docker compose up
 ```
 
+## Deployment
+
+Pushing to `main` runs `.github/workflows/pages.yml`, which rebuilds `site/data.json` and
+publishes `site/` to GitHub Pages.
+
 ## Roadmap
 
-- [x] Show paper title, abstract and PDF link
-- [x] Allow figure to stretch with screen size
-- [x] Cache embeddings to disk, read from disk
-- [x] Add links to the GitHub repo and my LinkedIn
-- [ ] Mark papers with code, add 'see code' button
-- [ ] Allow searching by keyword
-- [x] Use OpenAI embeddings
-- [x] Add more years, allow selecting year
-- [x] Add ECCV
-- [ ] Add ICCV
-- [ ] Docker image that builds on both arm and x86
+- [x] SPECTER2 embeddings (open model, no API key)
+- [x] 2D projection with clusters and automatic topic labels
+- [x] Static site — no server, no cold start
+- [x] Keyword search across titles and abstracts
+- [ ] Mark papers with released code
+- [ ] Add more conferences and years back

build_site.py

@@ -0,0 +1,47 @@
+"""Build the static site payload served by GitHub Pages.
+
+Merges data/cvpr_2026_papers.json and data/cvpr_2026_layout.json into a single
+column-oriented site/data.json, which site/index.html renders client-side with
+plotly.js. Run after the offline pipeline (scrape.py, embed.py, layout.py).
+"""
+import json
+import os
+
+PAPERS_PATH = "data/cvpr_2026_papers.json"
+LAYOUT_PATH = "data/cvpr_2026_layout.json"
+OUT_PATH = "site/data.json"
+
+
+def main():
+    with open(PAPERS_PATH) as f:
+        papers = json.load(f)
+    with open(LAYOUT_PATH) as f:
+        layout_data = json.load(f)
+
+    points = layout_data["points"]
+    assert len(points) == len(papers), (
+        f"point count {len(points)} != paper count {len(papers)}"
+    )
+
+    payload = {
+        "clusters": layout_data["clusters"],
+        # Column-oriented: ~15% smaller than a list of per-paper objects.
+        "x": [round(p["x"], 3) for p in points],
+        "y": [round(p["y"], 3) for p in points],
+        "c": [p["cluster"] for p in points],
+        "title": [p["title"] for p in papers],
+        "authors": [p["authors"] for p in papers],
+        "abstract": [p["abstract"] for p in papers],
+        "pdf": [p["pdf_link"] for p in papers],
+    }
+
+    os.makedirs("site", exist_ok=True)
+    with open(OUT_PATH, "w") as f:
+        json.dump(payload, f, separators=(",", ":"), ensure_ascii=False)
+
+    mb = os.path.getsize(OUT_PATH) / 1e6
+    print(f"wrote {OUT_PATH}: {len(papers)} papers, {mb:.1f} MB")
+
+
+if __name__ == "__main__":
+    main()

site/index.html

@@ -0,0 +1,190 @@
+<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>CVPR 2026 Explorer — SPECTER2 Semantic Map</title>
+<meta name="description" content="A 2D semantic map of all CVPR 2026 papers, built from SPECTER2 embeddings.">
+<script src="https://cdn.plot.ly/plotly-gl2d-2.35.2.min.js" charset="utf-8"></script>
+<style>
+  :root { color-scheme: dark; }
+  * { box-sizing: border-box; }
+  body {
+    margin: 0; background: #111; color: #eee;
+    font-family: -apple-system, BlinkMacSystemFont, "Segoe UI", Roboto, Helvetica, Arial, sans-serif;
+  }
+  header {
+    display: flex; flex-wrap: wrap; gap: 12px; align-items: center;
+    padding: 12px 16px; border-bottom: 1px solid #2a2a2a;
+  }
+  h1 { font-size: 18px; margin: 0; font-weight: 600; }
+  h1 small { color: #888; font-weight: 400; }
+  #search {
+    flex: 1 1 220px; max-width: 380px; padding: 7px 10px;
+    background: #1c1c1c; color: #eee; border: 1px solid #3a3a3a; border-radius: 6px;
+    font-size: 14px;
+  }
+  #search:focus { outline: none; border-color: #6ea8fe; }
+  #hits { color: #888; font-size: 13px; min-width: 90px; }
+  header a { color: #9ab; font-size: 13px; text-decoration: none; }
+  header a:hover { text-decoration: underline; }
+  main { display: flex; flex-wrap: wrap; }
+  #plot { flex: 1 1 620px; height: calc(100vh - 130px); min-height: 420px; }
+  #panel {
+    flex: 0 1 380px; max-height: calc(100vh - 130px); overflow-y: auto;
+    padding: 16px; border-left: 1px solid #2a2a2a; font-size: 14px; line-height: 1.5;
+  }
+  #panel h2 { font-size: 16px; margin: 0 0 8px; }
+  #panel .authors { color: #9a9a9a; font-size: 13px; margin-bottom: 12px; }
+  #panel .abstract { color: #ccc; }
+  #panel a.pdf {
+    display: inline-block; margin-bottom: 12px; padding: 6px 12px;
+    background: #2b6cb0; color: #fff; border-radius: 6px; text-decoration: none; font-size: 13px;
+  }
+  #panel .hint { color: #777; }
+  footer { padding: 10px 16px; color: #777; font-size: 12px; border-top: 1px solid #2a2a2a; }
+  footer a { color: #9ab; }
+  @media (max-width: 800px) {
+    #plot { height: 60vh; }
+    #panel { border-left: none; border-top: 1px solid #2a2a2a; max-height: none; }
+  }
+</style>
+</head>
+<body>
+<header>
+  <h1>CVPR 2026 Explorer <small>— SPECTER2 semantic map</small></h1>
+  <input id="search" type="search" placeholder="Search titles &amp; abstracts…" autocomplete="off" disabled>
+  <span id="hits"></span>
+  <a href="https://github.com/flecomet/cvpr-explorer" target="_blank" rel="noopener">GitHub</a>
+</header>
+<main>
+  <div id="plot"></div>
+  <aside id="panel"><p class="hint">Loading 4000+ papers…</p></aside>
+</main>
+<footer>
+  Fork of <a href="https://github.com/dataplayer12/cvpr-explorer" target="_blank" rel="noopener">dataplayer12/cvpr-explorer</a>.
+  Nearby points are semantically similar papers. Click a point for details.
+</footer>
+
+<script>
+const panel = document.getElementById("panel");
+const search = document.getElementById("search");
+const hits = document.getElementById("hits");
+let D = null, traces = null, haystack = null;
+
+// Golden-angle hue rotation: distinct colors for an arbitrary cluster count.
+function clusterColor(i) {
+  return "hsl(" + ((i * 137.508) % 360).toFixed(1) + ",65%,60%)";
+}
+
+function buildTraces(data) {
+  const byCluster = new Map();
+  data.c.forEach((cid, i) => {
+    if (!byCluster.has(cid)) byCluster.set(cid, []);
+    byCluster.get(cid).push(i);
+  });
+  // -1 ("unclustered") first so it renders underneath everything else.
+  const ids = [...byCluster.keys()].sort((a, b) => (a === -1 ? -1 : b === -1 ? 1 : a - b));
+  return ids.map((cid, n) => ({
+    type: "scattergl",
+    mode: "markers",
+    // Labels are "term · term · term"; the first two keep the legend narrow.
+    name: data.clusters[String(cid)].split(" · ").slice(0, 2).join(" · "),
+    x: byCluster.get(cid).map(i => data.x[i]),
+    y: byCluster.get(cid).map(i => data.y[i]),
+    text: byCluster.get(cid).map(i => data.title[i]),
+    customdata: byCluster.get(cid),
+    marker: { size: 6, color: cid === -1 ? "#666" : clusterColor(n) },
+    hovertemplate: "<b>%{text}</b><extra></extra>",
+  }));
+}
+
+const LAYOUT = {
+  margin: { l: 40, r: 20, b: 40, t: 20 },
+  paper_bgcolor: "#111", plot_bgcolor: "#111",
+  font: { color: "#eee" },
+  xaxis: { showgrid: false, zeroline: false, showticklabels: false },
+  // scaleanchor keeps the map's true aspect ratio: distance means the same on both axes.
+  yaxis: { showgrid: false, zeroline: false, showticklabels: false, scaleanchor: "x" },
+  legend: { font: { size: 10 }, itemsizing: "constant", itemwidth: 30 },
+  dragmode: "pan",
+  hovermode: "closest",
+};
+
+function showPaper(i) {
+  panel.innerHTML = "";
+  const h2 = document.createElement("h2");
+  h2.textContent = D.title[i];
+  const au = document.createElement("p");
+  au.className = "authors";
+  au.textContent = D.authors[i];
+  const a = document.createElement("a");
+  a.className = "pdf";
+  a.href = D.pdf[i];
+  a.target = "_blank";
+  a.rel = "noopener";
+  a.textContent = "Open PDF";
+  const ab = document.createElement("p");
+  ab.className = "abstract";
+  ab.textContent = D.abstract[i];
+  panel.append(h2, au, a, ab);
+  panel.scrollTop = 0;
+}
+
+// Search dims non-matching points instead of hiding them, so the map keeps its shape.
+function applySearch(q) {
+  q = q.trim().toLowerCase();
+  if (!q) {
+    Plotly.restyle("plot", { "marker.opacity": 1, selectedpoints: [null] });
+    hits.textContent = "";
+    return;
+  }
+  let n = 0;
+  const selected = traces.map(t => {
+    const sel = [];
+    t.customdata.forEach((i, k) => { if (haystack[i].includes(q)) { sel.push(k); n++; } });
+    return sel;
+  });
+  Plotly.restyle("plot", { selectedpoints: selected, "marker.opacity": 1 });
+  Plotly.restyle("plot", {
+    "selected.marker.opacity": 1,
+    "unselected.marker.opacity": 0.06,
+  });
+  hits.textContent = n + " match" + (n === 1 ? "" : "es");
+}
+
+fetch("data.json").then(r => r.json()).then(data => {
+  D = data;
+  traces = buildTraces(data);
+  haystack = data.title.map((t, i) => (t + " " + data.abstract[i]).toLowerCase());
+  Plotly.newPlot("plot", traces, LAYOUT, {
+    responsive: true, displaylogo: false, scrollZoom: true,
+    modeBarButtonsToRemove: ["select2d", "lasso2d", "toggleSpikelines"],
+  });
+  document.getElementById("plot").on("plotly_click", e => showPaper(e.points[0].customdata));
+  panel.innerHTML = '<p class="hint">A 2D semantic map of all ' + data.x.length +
+    " CVPR 2026 papers, built from SPECTER2 embeddings.<br><br>" +
+    "Click any point to read its title, authors and abstract, and to open the PDF. " +
+    "Scroll to zoom, drag to pan.</p>";
+  search.disabled = false;
+  let timer;
+  search.addEventListener("input", () => {
+    clearTimeout(timer);
+    timer = setTimeout(() => {
+      applySearch(search.value);
+      // Keep the URL shareable without adding history entries per keystroke.
+      const url = new URL(location);
+      if (search.value.trim()) url.searchParams.set("q", search.value);
+      else url.searchParams.delete("q");
+      history.replaceState(null, "", url);
+    }, 150);
+  });
+
+  const q = new URLSearchParams(location.search).get("q");
+  if (q) { search.value = q; applySearch(q); }
+}).catch(err => {
+  panel.innerHTML = '<p class="hint">Failed to load data.json: ' + err + "</p>";
+});
+</script>
+</body>
+</html>