Compare reference and distorted inputs across scoring backends and report pairwise numerical deltas.
npx skills add https://github.com/VMAFx/vmafx --skill validate-scores
Please help me install this Agent Skill. Skill: validate-scores Source: https://github.com/VMAFx/vmafx/tree/main/.claude/skills/validate-scores Command: npx skills add https://github.com/VMAFx/vmafx --skill validate-scores
This Skill removes the manual effort of comparing model scores across multiple backends by automating parity checks, score aggregation, and numerical difference reporting.
Ask the assistant to validate a reference and distorted clip pair across the selected backends and report any score differences that exceed the allowed ULP threshold.
High-intent search queries and answers about installing and using this skill.
Backend parity for VMAF score comparison is verified by running identical reference and distorted inputs through every enabled backend, computing pairwise absolute differences and ULP distances, and applying a two-ULP pass or fail gate to detect floating-point drift.
ULP distance measurement in floating-point score validation quantifies the numerical drift between different backend implementations, determining whether per-frame score outputs remain within an accepted tolerance budget of two ULPs for bit-exactness verification.
Automate regression testing for a new SIMD path by parsing per-frame JSON scores from identical inputs across backends, calculating max absolute differences, and reporting any numerical deltas that exceed the allowed ULP threshold before merging.
Bit-exactness verification for multiple scoring backends is needed during hot-path changes, SIMD path updates, or GPU backend modifications to prove that per-feature score outputs remain within the accepted tolerance budget without manual score aggregation.
Yes, backend score validation operates with no external dependencies, handling per-frame JSON score parsing, max absolute difference calculation, and ULP distance measurement internally to report pairwise numerical deltas across enabled backends.
The best way to check floating-point drift between scalar and GPU backend implementations is to compare identical reference and distorted inputs across all enabled backends, measuring pairwise ULP distances against a strict two-ULP pass or fail gate.
VMAFxDetect cross-backend VMAF score drift using per-frame JSON and ULP distance.
swarm-ai-researchAggregate fresh RSS feed items and generate concise summaries.
Arcadi4Multiply polynomials via FFT with padding and root-of-unity handling.
SolidLaoSelect aggregation algorithms for SQL GROUP BY queries based on cardinality and memory.
