tao-train-nvpanoptix3d

NVPanoptix3D for panoptic 3D scene reconstruction from posed RGB images. Produces 3D panoptic segmentation (semantic, instance, and panoptic masks) with occupancy completion. Built on a VGGT backbone with a Mask2Former-style head and 3D frustum reconstruction. Use when training, evaluating, exporting, or running inference for a TAO NVPanoptix3D model. Trigger phrases include "train NVPanoptix3D", "panoptic 3D reconstruction", "3D scene segmentation", "occupancy completion".

Quality

88%

Does it follow best practices?

Run evals on this skill

Adds up to 20 points to the overall score

View guide

Securityby

Low

Low-risk findings worth noting

Quality

Content

88%Weight 40%Scale 1-5

Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.

The body is highly actionable with executable commands, concrete spec overrides, and explicit validation/feedback loops for batch and resume operations. Its main weakness is redundancy with the packaged YAML references, which slightly undercuts conciseness and progressive disclosure.

Suggestions

Replace the inlined Per-Action Dataset Requirements and Spec Param handoff tables with concise pointers to references/skill_info.yaml (data_sources and spec_params), keeping only the action-specific gotchas inline, to reduce duplication and improve conciseness.

Trim the opening descriptive paragraph that repeats the frontmatter description; the body can assume that context and jump straight to the Quick Start.

Add a short 'References' index section listing each bundle file (skill_info.yaml, spec_template_train/evaluate/export/inference.yaml) with a one-line purpose so navigation is fully explicit.

Dimension	Reasoning	Score
Conciseness	The body is dense and operational with little concept padding, but several tables (Per-Action Dataset Requirements, Spec Param handoff mappings) duplicate content already in references/skill_info.yaml, and the opening description repeats the frontmatter, leaving minor trim opportunities that keep it just below 5.	4 / 5
Actionability	Copy-paste-ready docker run commands for train/evaluate/inference/export, concrete spec_override Python dicts, exact module entrypoint paths, and specific error-pattern fixes make the guidance fully executable across common cases.	5 / 5
Workflow Clarity	Clear per-action command sequences plus explicit validation checkpoints and feedback loops (verify the promoted AutoML job resumed from the explicit epoch/step checkpoint and produced a usable checkpoint; the resume-epoch-boundary guidance on setting num_epochs two epochs beyond the smoke run) satisfy the destructive/batch validation expectation.	5 / 5
Progressive Disclosure	The body points to real one-level-deep bundle files (references/skill_info.yaml, references/spec_template_<action>.yaml) by path with clear signaling, but it also inlines material that mirrors those YAMLs (dataset table, spec_param mappings), leaving minor organization gaps versus a clean 5.	4 / 5
	Total	18 / 20 Passed

Description

87%Weight 40%Scale 1-5

Based on the skill's description, can an agent find and select it at the right time? Clear, specific descriptions lead to better discovery.

The description is specific, well-triggered, and distinctive, explicitly answering both what the skill does and when to use it with concrete trigger phrases. Minor room remains to broaden trigger-term synonyms and sharpen the action verbs beyond generic TAO stages.

Dimension	Reasoning	Score
Specificity	Lists multiple concrete outputs ("3D panoptic segmentation (semantic, instance, and panoptic masks) with occupancy completion") plus architecture detail (VGGT backbone, Mask2Former-style head, 3D frustum), but the train/eval/export/inference verbs are generic TAO actions rather than domain-specific actions, so it stops short of a clean 5.	4 / 5
Completeness	Clearly states what (panoptic 3D scene reconstruction producing semantic/instance/panoptic masks with occupancy completion) and explicitly when ("Use when training, evaluating, exporting, or running inference for a TAO NVPanoptix3D model") with concrete trigger phrases, matching the 5 anchor.	5 / 5
Trigger Term Quality	Explicit trigger phrases ("train NVPanoptix3D", "panoptic 3D reconstruction", "3D scene segmentation", "occupancy completion") give good natural keyword coverage a user would say, but lack synonyms/file-extension variants that would push it to 5.	4 / 5
Distinctiveness Conflict Risk	The narrow model name NVPanoptix3D and specific panoptic-3D-reconstruction niche give a clear trigger surface with minimal overlap risk against other skills.	5 / 5
	Total	18 / 20 Passed

Validation

87%

Warnings & errors only

Checks the skill against the spec for correct structure and formatting. All validation checks must pass before discovery and implementation can be scored.

Validation — 14 / 16 Passed

Validation for skill structure

Criteria	Description	Result
frontmatter_unknown_keys	Unknown frontmatter key(s) found; consider removing or moving to metadata	Warning
referenced_paths_exist	Referenced path issues: 2 missing	Warning

	Total	14 / 16 Passed

Repository: NVIDIA/skills
Path: skills/tao-train-nvpanoptix3d/SKILL.md
Commit: f436f05

Reviewed: 4 days ago

Table of Contents

Discovery Implementation Validation

Is this your skill?

If you maintain this skill, you can claim it as your own. Once claimed, you can manage eval scenarios, bundle related skills, attach documentation or rules, and ensure cross-agent compatibility.