Spaces:
Running
Running
| title: World Memory Deck | |
| emoji: π§ | |
| colorFrom: green | |
| colorTo: indigo | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| license: mit | |
| # World Memory β Presentation | |
| A reveal.js slide deck for **World Memory: Consistent Long-Video Generation via | |
| Knownness-Conditioned Diffusion Forcing** (working title, CoRL 2026). | |
| The deck walks through the full pipeline: | |
| - **Phase 1 β Offline caching:** posed video β persistent 3D `PointMemory` β | |
| reprojected `z_mem`, per-region `knownness k`, per-token 3D address `Xtok`, | |
| and a compact memory slot bank. | |
| - **Phase 2 β Model:** a frozen Wan Video DiT with four small zero-init hooks | |
| (per-token noise from `k`, 3D positional encoding, geometry-keyed cross-attention | |
| with Rope3D, confidence-gated shortcut). | |
| - **Phase 3 β Training:** anchor-mixture Diffusion Forcing β repair memory, never | |
| copy it; the target is always the real frame. | |
| - **Phase 4 β Inference & rollout:** knownness sets a per-token start-noise level; | |
| known regions reuse-and-repair, unknown regions generate, and new frames are | |
| written back into the world memory for consistent revisits. | |
| ## View | |
| Open the Space β `index.html` renders the deck directly. Navigate with arrow keys, | |
| press <kbd>F</kbd> for fullscreen, <kbd>S</kbd> for speaker notes, <kbd>O</kbd> for | |
| the slide overview, and <kbd>Esc</kbd> to zoom out. | |
| Built with [reveal.js](https://revealjs.com) (loaded from CDN). The centerpiece | |
| pipeline figure is a hand-authored inline SVG. | |