nvidia/synthda-demo

1---2title: SynthDA Motion Playground3emoji: 🟢4colorFrom: green5colorTo: gray6sdk: static7app_file: index.html8pinned: false9short_description: Explore human motion, pose variations and original videos10license: fair-noncommercial-research-license11tags:12  - human-motion13  - synthetic-data14  - action-recognition15---16 17# SynthDA Motion Playground18 19An interactive viewer for **actual saved SynthDA poses**. Compare two input20motions with an augmented output, inspect a lineup of five variations, and21follow a source RGB clip through pose extraction and a fitted SOMA human visualization.22 23[Open the demo](https://huggingface.co/spaces/nvidia/synthda-demo) ·24[Research code and publications](https://github.com/NVIDIA/synthda)25 26![Motion playground](assets/social-preview.png)27 28## Included in this release29 30- **Video → extracted pose → human figure:** one source video drives both31  the extracted skeleton and a fitted native SOMA body. Toggle the body and joint overlay.32- **Motion playground:** Compare motions, Variation lineup and Overlay poses;33  saved-variant selection, full camera rotation, zoom, frame scrubbing, playback34  speed, SOMA/Skeleton display, Original/Smoothed preview and original-pose download.35- **Variation gallery:** three verified fall pose examples, two short motion36  studies, and all 33 videos from the original Hugging Face Space.37- **SOMA bodies:** 33 fitted motions (six sources and 27 variants), rendered locally38  in WebGL 2 with the native SOMA-X neutral body. Hands stay neutral; fitted surfaces39  approximate the saved joints and do not reconstruct the original person.40- Additional examples and sourced recognition-result records can be added through the manifest.41 42This release contains **one action class: falling**. Multiple fall examples are43not represented as multiple action classes. There are no fabricated recognition44scores or motion-quality measurements. This is saved-result visualization,45not a live inference endpoint.46 47## Run locally48 49No Node install, build step, GPU or backend is required for the viewer.50 51```sh52python3 -m http.server 7860 --bind 127.0.0.153```54 55Open http://localhost:7860. Serve the folder over HTTP; opening `index.html`56directly from the filesystem will block JSON loading in many browsers.57 58## Publish or extend59 60See [DEPLOYMENT.md](DEPLOYMENT.md) for Hugging Face and GitHub instructions,61[ASSET_FORMAT.md](ASSET_FORMAT.md) for additional examples and meshes, and62[ASSET_PROVENANCE.md](ASSET_PROVENANCE.md) for source revisions and limitations.63 64The Space's pre-existing license metadata is preserved above. Upstream65repository material retains its [NSCLv2 notice](UPSTREAM-LICENSE.md); this66viewer does not relicense upstream assets. Publicly hosting saved visualizations67does not change the license terms of the generation pipeline or source assets.68 69The SOMA-derived assets retain their [Apache-2.0 license](assets/soma/LICENSE.txt)70and [notice](assets/soma/NOTICE.txt). Three.js retains its [MIT license](vendor/THREE-LICENSE.txt).71See [SOMA.md](SOMA.md) for fitting, rendering and reproduction details.72