Sovereign 8s.is mirror of standardgalactic/FAME
  • CSS 50.4%
  • HTML 34.6%
  • JavaScript 15%
Find a file
Repository files (latest commit first)
Filename Latest commit message Latest commit date
2026-09-26 12:54:44 +08:00
assets Update teaser captions, evaluation screenshot, and double-speed looping videos 2026-09-22 16:01:47 +08:00
static Apply Rome-style red-orange gradient to Spotlight parentheses 2026-09-26 12:54:44 +08:00
.nojekyll Publish All Roads Lead to Rome project website with updated low-angle teaser 2026-09-22 01:17:05 +08:00
app.js Add files via upload 2026-09-20 20:35:04 +08:00
index.html Apply Rome-style red-orange gradient to Spotlight parentheses 2026-09-26 12:54:44 +08:00
README.md Update public website address to FAME 2026-09-22 16:29:48 +08:00
styles.css Add files via upload 2026-09-20 20:32:09 +08:00

All Roads Lead to Rome — Project page

Self-contained static project page. Open index.html to preview. No build step or package installation is needed.

All four videos start muted automatically at 2× speed when the page opens and loop continuously. Native controls remain available if browser-level autoplay restrictions require a manual play action. The custom pair-control buttons have been removed.

GitHub Pages

Published website: https://vollichor.github.io/FAME/

Repository: https://github.com/vollichor/FAME

  1. Upload the contents of this folder to the root of a GitHub repository. Keep index.html, assets, and static at the same level.
  2. In the repository, open Settings → Pages.
  3. Under Build and deployment, choose Deploy from a branch, then main and / (root), and save.
  4. Use the website address shown on the Pages settings screen after deployment.

The ZIP contains the complete deployable website. _qa is a local validation directory and is excluded from the ZIP.

Sources and editorial choices

  • Paper title, method descriptions and Tables 1–3: supplied 4774_All_Roads_Lead_to_Rome_Fl.pdf.
  • Figures: high-resolution crops of Figures 1–3, plus original embedded images from NeurIPS2026-demo.pptx. PNG exports retain sharp text. SVG preserves source vectors and embedded images, not vector detail that did not exist in the source.
  • Website author names and affiliations were supplied by the user: Yang Li (1), Aming WU (2), Zihao Zhang (1), Ziju Han (1), Sijia Zhang (1), Yahong Han (1). Affiliation 1 is Tianjin University; affiliation 2 is Hefei University of Technology. The supplied PDF is unchanged, and the page continues to label the document as a NeurIPS 2026 submission.
  • Benchmark numbers reproduce the manuscript and are not new independently validated experiments.
  • The selected videos are existing NextBestPath Simple/Hard checkpoint runs. They are not a validated comparison between NextBestPath and the paper's proposed flow-driven method. The requested visible column labels are presentation labels; actual checkpoint provenance is documented here. Replace those files when genuine method-comparison runs are available.
  • The Evaluation Results section displays the user-supplied original screenshot, assets/images/evaluation-results-tables.png, without resampling.
  • butcher_in_the_sandstorm_15 uses the updated centered-view videos.
  • No analytics, trackers or third-party scripts are included. External project links open only when clicked.

Layout acknowledgments

Layout and style follow https://shiyao-li.github.io/nbp/ and the Nerfies project page template: https://github.com/nerfies/nerfies.github.io . Bulma and Bootstrap stylesheet headers retain their upstream license notices.

Media

The homepage teaser uses the 5120×1480 lossless PNG assets/images/fear_in_the_abyss_10_low_view_no_text_5K.png. Left: a real Simple-trained NextBestPath run on the Hard scene fear_in_the_abyss_10, stopped at 70.04% measured coverage after a nominal 1,104 steps, without manual guidance. Right: the complete Ground Truth geometry with recorded and illustrative orthogonal room routes, not a method reconstruction. The panels share a camera and scale; ceiling points are removed for visualization only. This is an illustrative comparison with reference geometry, not an evaluation of the proposed flow-driven method. The requested visible captions are presentation labels, not evidence of method performance; the image alternative text preserves the actual provenance. Click the homepage image to open the 5K original.

Each row contains Simple-trained NBP on the left and Hard-trained NBP on the right. Both source videos are 1920×1080, 16 fps and 25 seconds, and use a 400-step exploration budget. At 2× speed, one loop takes approximately 12.5 seconds. Playback, pause and seeking remain synchronized within each pair through the native controls.