Physical AI Video Data Augmentation
暂无简介
- License 许可
- MIT
- 所属仓库
- rhCat/skillChip
- 最近更新
- 2026-07-31
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- Physical AI Video Data Augmentation
- perks
- [ "run", "gen-configs", "prepare-assets", "render-compare" ]
- skill
- physical-ai-video-data-augmentation
02 资源文件 48
- 其他blueprint.drawio2.7 KB
- 参考文档blueprint.json3.6 KB
- 资产文件blueprint.svg8.9 KB
- 参考文档index.json5.3 KB
- 参考文档ledger.json451 B
- 参考文档perks.json1.0 KB
- 参考文档perks/gen-configs/manifesto.json417 B
- 参考文档perks/gen-configs/metadata.json1.4 KB
- 参考文档perks/gen-configs/src/contracts.json421 B
- 脚本perks/gen-configs/src/generate_configs.py6.2 KB
- 脚本perks/gen-configs/src/generate_configs.sh1.0 KB
- 参考文档perks/gen-configs/test/case.json308 B
- 参考文档perks/gen-configs/test/fixture/cfg/augmentation/augmentation.yaml117 B
- 参考文档perks/gen-configs/test/fixture/cfg/workflow_config.yaml179 B
- 资产文件perks/gen-configs/test/fixture/videos/clip.mp411 B
- 参考文档perks/prepare-assets/manifesto.json360 B
- 参考文档perks/prepare-assets/metadata.json1.3 KB
- 参考文档perks/prepare-assets/src/contracts.json331 B
- 脚本perks/prepare-assets/src/prepare_demo_assets.impl.sh5.5 KB
- 脚本perks/prepare-assets/src/prepare_demo_assets.sh853 B
- 参考文档perks/prepare-assets/test/case.json260 B
- 其他perks/prepare-assets/test/fixture/.gitkeep—
- 参考文档perks/render-compare/manifesto.json483 B
- 参考文档perks/render-compare/metadata.json1.4 KB
- 参考文档perks/render-compare/src/contracts.json473 B
- 脚本perks/render-compare/src/render_side_by_side.impl.sh2.1 KB
- 脚本perks/render-compare/src/render_side_by_side.sh1.1 KB
- 参考文档perks/render-compare/test/case.json649 B
- 其他perks/render-compare/test/fixture/.gitkeep—
- 参考文档perks/run/manifesto.json509 B
- 参考文档perks/run/metadata.json1.5 KB
- 参考文档perks/run/src/contracts.json340 B
- 脚本perks/run/src/cosmos_worker.sh5.6 KB
- 脚本perks/run/src/endpoint_common.sh7.6 KB
- 脚本perks/run/src/generate_configs.py6.2 KB
- 脚本perks/run/src/llm_server.sh501 B
- 脚本perks/run/src/osmo_barrier.py5.2 KB
- 脚本perks/run/src/pl_augmented_worker.sh5.2 KB
- 脚本perks/run/src/pl_original_worker.sh5.8 KB
- 脚本perks/run/src/pre_submit_guard.py9.2 KB
- 脚本perks/run/src/pre_submit_guard.sh743 B
- 脚本perks/run/src/preflight_credentials.impl.sh20.9 KB
- 脚本perks/run/src/preflight_credentials.sh1.2 KB
- 脚本perks/run/src/prepare_demo_assets.sh5.5 KB
- 脚本perks/run/src/render_side_by_side.sh2.1 KB
- 脚本perks/run/src/stage_run_artifacts.sh2.6 KB
- 脚本perks/run/src/vlm_server.sh960 B
- 参考文档perks/run/test/case.json221 B
03 SKILL.md 正文
physical-ai-video-data-augmentation — Physical AI Video Data Augmentation
Orchestrate the end-to-end video data augmentation + auto-labeling workflow on OSMO — flow selection, credential preflight, submit-time guard, GPU submit, monitoring, and output retrieval.
Perks
| perk | tool(s) | nature |
|---|---|---|
run |
preflight_credentials, pre_submit_guard |
destructive (deploys/repairs in-cluster NIMs, submits GPU workflows to OSMO, writes OSMO credentials) |
gen-configs |
generate_configs |
read-only inputs (deterministic per-video augmentation configs + manifest under record_store) |
prepare-assets |
prepare_demo_assets |
pulls the VDA demo dataset from Hugging Face and flattens it into one demo input dir |
render-compare |
render_side_by_side |
read-only inputs (side-by-side input-vs-augmented MP4 from a staged local run) |
The run perk gates a VDA submit: preflight_credentials validates secrets, OSMO credentials, the
control plane (ONLINE GPU pool + GPU POD_TEMPLATE), and workflow image access, then pre_submit_guard
validates the rendered workflow YAML (setup.files coverage, dataset/cache URLs, VIDEO_NAME basenames)
before any osmo workflow submit. Because the workflow deploys NIMs and submits GPU jobs to OSMO, the
perk is declared destructive: true; the executor gates it accordingly. Each tool emits one line of
structured JSON and writes its report under record_store.
The other three perks are standalone, non-destructive operations that bracket a run:
gen-configs runs generate_configs.py to deterministically render one Cosmos augmentation config per
input video (plus a manifest.yaml) from a pipeline config dir; prepare-assets runs
prepare_demo_assets.sh to pull the nvidia/video-data-augmentation-demo videos from Hugging Face and
flatten them into one demo input directory; and render-compare runs render_side_by_side.sh to stitch
the original input and the augmented output of a locally-staged run into one side-by-side MP4. Each emits
one structured-JSON line and writes under record_store.
How to use it
Copy ledger.json -> task-ledger.json, set the vars (WORKFLOW) + record_store, then
validate -> compose -> compile -> oversight -> executor.
Localized from NVIDIA/skills
physical-ai-video-data-augmentation(Apache-2.0).
04 同仓库更多 Skills
Vercel Deploy
暂无简介
Audio Transcribe
暂无简介
Speech (OpenAI TTS)
暂无简介
Skill Installer
暂无简介
demo-skill
A demo skill used as a test fixture for the validate perk. Use when checking that quick_validate.py accepts a well-formed SKILL.md.
demo-skill
A demo skill used as a test fixture for generating agents/openai.yaml. Use when validating the gen_openai_yaml perk.