Spaces:
Running
Running
| <html lang="en"><head><meta charset="utf-8"><meta name="viewport" content="width=device-width,initial-scale=1"><title>SynthDA · Motion Playground</title><meta name="description" content="Explore real and generated motions, inspect saved SynthDA variations, and follow an original video through pose-level augmentation."><meta property="og:title" content="SynthDA · One action. A spectrum of variations."><meta property="og:description" content="Compare motions, explore pose variations and discover the original video gallery."><meta property="og:type" content="website"><meta property="og:image" content="https://huggingface.co/spaces/nvidia/synthda-demo/resolve/main/assets/social-preview.png"><meta name="twitter:card" content="summary"><link rel="stylesheet" href="style.css"><script type="module" src="app.js"></script><link rel="icon" href="data:image/svg+xml,%3Csvg xmlns=%27http://www.w3.org/2000/svg%27 viewBox=%270 0 32 32%27%3E%3Ccircle cx=%2716%27 cy=%2716%27 r=%2713%27 fill=%27%2376b900%27/%3E%3C/svg%3E"></head><body> | |
| <header class="masthead"><a class="brand" href="#playground"><span class="brand-dot"></span>SynthDA</a><span class="project-label">An NVIDIA research project</span><a class="source-link" href="https://github.com/NVIDIA/synthda" target="_blank" rel="noopener">Explore the research ↗</a></header> | |
| <main><section class="hero"><p class="eyebrow">Pose-level synthetic data augmentation</p><h1>One action.<br>A spectrum of variations.</h1><p class="lede">Explore how real and generated human motions become new training examples.</p><div class="hero-meta"><span>Real repository data</span><span>Interactive 3D poses</span><span>Saved results · no live inference</span></div></section> | |
| <nav class="sections" aria-label="Demo sections"><a href="#pipeline">01 <span>Video → extracted pose</span></a><a href="#playground" class="active">02 <span>Motion playground</span></a><a href="#gallery">03 <span>Variation gallery</span></a></nav> | |
| <section class="section" id="pipeline"><div class="section-heading"><div><p class="eyebrow">01 / From an existing example</p><h2>Follow the motion.</h2></div><p>See the pose extracted from one RGB video, then watch a human figure follow those same joints.</p></div><div class="pipeline-grid"><article><div class="step-label"><span>01</span><h3>Original video</h3></div><div class="video-well"><video id="pipeline-video" muted playsinline preload="metadata" aria-label="Original source video"></video></div><p id="pipeline-source">Source A</p></article><article><div class="step-label"><span>02</span><h3>Extracted pose</h3></div><canvas id="pipeline-pose" aria-label="Extracted 3D pose at the selected frame"></canvas><p>RGB → 3D joints</p></article><article><div class="step-label"><span>03</span><h3>Pose-driven human</h3></div><div class="soma-stage"><canvas id="pipeline-output" aria-label="Source pose skeleton"></canvas><canvas id="pipeline-soma" class="soma-layer" aria-label="SOMA body fitted to the extracted source pose"></canvas></div><p id="pipeline-variant">Same extracted pose · fitted SOMA body</p></article></div><div class="figure-controls"><label><input type="checkbox" id="pipeline-human" checked> SOMA human</label><label><input type="checkbox" id="pipeline-skeleton"> Show joints on figure</label><span>Pose visualization · not augmentation</span></div><div class="pipeline-controls"><button id="pipeline-play" type="button" disabled>Play sequence</button><label class="timeline-label" for="pipeline-frame">Sequence progress<input id="pipeline-frame" type="range" min="0" max="128" value="0" disabled></label><output id="pipeline-count">Frame 1 / 129</output></div><p class="caption">Uses source A of the example selected below. Both pose views show the same extracted motion; the variant slider does not change them. RGB follows normalized sequence progress, not verified event alignment.</p><p class="caption" id="video-error" role="status"></p></section> | |
| <section class="section" id="playground"><div class="section-heading"><div><p class="eyebrow">02 / Explore the actual poses</p><h2>Motion playground.</h2></div><label class="example-label" for="example">Motion example<select id="example" disabled><option>Loading examples…</option></select></label></div><div class="playground-toolbar"><div class="mode-buttons" role="group" aria-label="Comparison mode"><button data-mode="compare" aria-pressed="true" type="button">Compare motions</button><button data-mode="lineup" aria-pressed="false" type="button">Variation lineup</button><button data-mode="overlay" aria-pressed="false" type="button">Overlay poses</button></div><label class="mesh-control" hidden><input type="checkbox" id="show-mesh"> Mesh overlay</label><span id="action-label" class="small-label">Fall · saved poses</span></div><div class="smoothing-controls"><label for="body-view">Show <select id="body-view"><option value="soma" selected>SOMA body</option><option value="skeleton">Skeleton</option></select></label><label><input id="body-joints" type="checkbox"> Show pose joints</label><label for="smoothing">Motion preview <select id="smoothing"><option value="smooth" selected>Smoothed</option><option value="original">Original</option></select></label><span id="smoothing-note">Temporal smoothing for display · original data preserved</span></div><p id="soma-status" class="caption" role="status" aria-live="polite"></p><div class="motion-stage"><div id="motion-labels" class="motion-labels"></div><div class="soma-stage"><canvas id="motion-canvas" aria-label="Interactive 3D skeleton comparison">Canvas support is required to display the motions.</canvas><canvas id="motion-soma" class="soma-layer" aria-label="Interactive SOMA body comparison"></canvas></div><div class="stage-footer"><span id="prompt"></span><span>Drag to rotate · scroll to zoom</span></div></div><div class="transport"><button type="button" id="play" disabled>Play</button><input type="range" id="frame" min="0" max="128" value="0" aria-label="Pose frame" disabled><output id="frame-count">Frame 1 / 129</output><label class="speed-label">Speed<select id="speed"><option value="0.5">0.5×</option><option value="1" selected>1×</option><option value="1.5">1.5×</option></select></label></div><div class="motion-controls"><label for="variant"><span class="control-heading">Saved variant <output id="variant-value">wA 0.5 · wB 0.5</output></span><input id="variant" type="range" min="1" max="9" step="1" value="5"><span class="range-ends"><span>wA 0.1</span><span>wA 0.9</span></span></label><label for="camera"><span class="control-heading">Camera angle <output id="camera-value">20°</output></span><input id="camera" type="range" min="-180" max="180" step="1" value="20"><span class="range-ends"><span>−180°</span><span>Front 0°</span><span>+180°</span></span></label><label for="zoom"><span class="control-heading">Zoom <output id="zoom-value">1×</output></span><input id="zoom" type="range" min="0.6" max="1.6" step="0.05" value="1"><span class="range-ends"><span>Wide</span><span>Close</span></span></label></div><div class="playground-note"><p id="variant-note">The slider selects a saved optimizer output. Its parameters are not percentages of an endpoint crossfade.</p><button type="button" id="download-pose" disabled>Download original poses</button></div><details class="provenance"><summary>About these motions and their provenance</summary><div id="provenance"></div><p>The grid is a visual reference, not recovered ground geometry. Playback defaults to 25 fps for this preview. Joint coordinates retain the repository's root-relative representation.</p><p id="mesh-note">SOMA bodies are fitted visualizations of the saved joints, with neutral hands and a fixed body shape. They are not recovered subject geometry. Original pose downloads are unchanged.</p></details><p id="load-status" class="status" role="status" aria-live="polite">Loading verified motion examples…</p></section> | |
| <section class="section" id="gallery"><div class="section-heading"><div><p class="eyebrow">03 / Motion studies and original renders</p><h2>Variation gallery.</h2></div><p id="gallery-scope">Fall examples from the public release.</p></div><div id="example-cards" class="example-cards"></div><div class="study-grid"><article><video controls playsinline preload="metadata" src="assets/video/compare-study.mp4" aria-label="Source motions and augmented output motion study"></video><h3>Source motions → saved output</h3><p>A synchronized skeleton study of the first fall example.</p><a href="assets/video/compare-study.mp4" download>Download film ↗</a></article><article><video controls playsinline preload="metadata" src="assets/video/spectrum-study.mp4" aria-label="Five saved pose variations motion study"></video><h3>A lineup of possibilities</h3><p>Five saved outputs shown together on one stage.</p><a href="assets/video/spectrum-study.mp4" download>Download film ↗</a></article></div><div class="archive-heading"><div><h3>From the original Space</h3><p>All 33 published clips, preserved in their original collections.</p></div><label for="archive-set">Collection<select id="archive-set" disabled></select></label></div><div class="archive-view"><video id="archive-video" controls playsinline preload="metadata" aria-label="Selected clip from the original Hugging Face demo"></video><div class="archive-controls"><p class="eyebrow">Original rendered result</p><h3 id="archive-title">Collection 1</h3><label for="archive-variant">Saved clip <output id="archive-value">0.5</output><input id="archive-variant" type="range" min="0" max="10" value="5" step="1"></label><p class="caption">Clip 0.0 and 1.0 are the original endpoint files. Source roles and correspondence to the playground examples have not been verified.</p><a id="archive-download" href="#" download>Download selected clip ↗</a><p id="archive-status" role="status" class="caption"></p></div></div><div id="recognition-results" hidden></div></section> | |
| <section class="closing"><p class="eyebrow">Explore the research</p><h2>Make more of the motion you already have.</h2><p>SynthDA explores pose-level augmentation for human action recognition, including underrepresented action classes.</p><a class="primary-link" href="https://github.com/NVIDIA/synthda" target="_blank" rel="noopener">Code, pipeline & publications ↗</a></section></main><footer class="page-footer"><span>SynthDA · Research demonstration</span><a href="UPSTREAM-LICENSE.md">Upstream license</a><a href="ASSET_PROVENANCE.md">Data & rendering notes</a><a href="https://huggingface.co/spaces/nvidia/synthda-demo" target="_blank" rel="noopener">Hugging Face ↗</a><a href="https://github.com/NVlabs/SOMA-X" target="_blank" rel="noopener">Body model: SOMA-X ↗</a></footer><noscript><p class="noscript">Enable JavaScript to use the motion controls. The study videos remain available in the gallery.</p></noscript></body></html> | |