Imported from nvidia-isaac/video_to_data (.claude/skills/run-ego-reconstruction-video/SKILL.md). Install upstream with npx skills add nvidia-isaac/video_to_data --skill run-ego-reconstruction-video. Copyright stays with the author.

Run Ego Reconstruction Video

Name: run-ego-reconstruction-video
Author: nvidia-isaac

Run from reconstruction/. Prefer the consolidated entrypoint and use HaMeR unless the user explicitly needs legacy DynHaMR:

python modules/v2d_pipelines/run_ego_reconstruction.py \
  --video <video.mp4> \
  --output_dir data/outputs/<run> \
  --object_prompt "<object>" \
  --hand_tracking hamer \
  --reference_frame 0

Add --object_mesh <mesh.obj> --skip_object_scale_estimation for a supplied mesh. Add --undistort, --run_droid_slam, --run_gravity_alignment, --run_gsplat_refinement, and --export_threejs_result only when requested.

Re-run the same command to resume cached stages. Confirm result/result.npz and result/mesh.obj; with post-processing, use the final suffixed result directory. Use the setup skill first when images, weights, Docker, or GPU prerequisites are missing.

{ "ocm": "1", "id": "nvidia-isaac-video-to-data-run-ego-reconstruction-video", "kind": "skill", "name": "run-ego-reconstruction-video", "description": "Run this repository’s egocentric reconstruction pipeline on a video. Use when a user asks to process an MP4 with HaMeR or DynHaMR hand tracking, an object prompt or mesh, undistortion, DROID-SLAM, gravity alignment, gsplat refinement, or a Three.js result export.", "publisher": "nvidia-isaac", "version": "1.0.0", "capabilities": { "domains": [ "general" ], "tags": [ "skill-md", "github" ], "languages": [ "en" ] }, "quality_prior": 0.6, "examples": [ "Run this repository’s egocentric reconstruction pipeline on a video. Use when a user asks to process an MP4 with HaMeR or DynHaMR hand tracking, an object prompt or mesh, undistortion, DROID-SLAM, gravity alignment, gsplat refinement, or a Three.js result export." ], "primary": false, "metadata": { "source": { "provider": "github", "repository": "https://github.com/nvidia-isaac/video_to_data", "path": ".claude/skills/run-ego-reconstruction-video/SKILL.md", "ref": "0b1d0f1771c5d49e31da417b0aa161bf8129d3ff", "url": "https://github.com/nvidia-isaac/video_to_data/blob/0b1d0f1771c5d49e31da417b0aa161bf8129d3ff/.claude/skills/run-ego-reconstruction-video/SKILL.md", "key": "nvidia-isaac/video_to_data/.claude/skills/run-ego-reconstruction-video/SKILL.md" } }, "instructions": "# Run Ego Reconstruction Video\n\nRun from `reconstruction/`. Prefer the consolidated entrypoint and use HaMeR\nunless the user explicitly needs legacy DynHaMR:\n\n```bash\npython modules/v2d_pipelines/run_ego_reconstruction.py \\\n --video <video.mp4> \\\n --output_dir data/outputs/<run> \\\n --object_prompt \"<object>\" \\\n --hand_tracking hamer \\\n --reference_frame 0\n```\n\nAdd `--object_mesh <mesh.obj> --skip_object_scale_estimation` for a supplied\nmesh. Add `--undistort`, `--run_droid_slam`, `--run_gravity_alignment`,\n`--run_gsplat_refinement`, and `--export_threejs_result` only when requested.\n\nRe-r", "cost": { "context_tokens": 212 } }

run-ego-reconstruction-video

About

Run Ego Reconstruction Video

Use it

Manifest

Reviews