What problem does it solve? Replacing a default model in an RDK Model Zoo sample with a self-trained or modified model breaks assumptions about class counts, shapes, layouts, preprocessing, and output decoding. This Skill aligns the model artifact, runtime, and sample code so the integration is verified rather than assumed. ## Core Features & Use Cases - I/O Contract Mapping: Builds a per-input and per-output comparison between the source model, compiled artifact metadata, and the sample wrapper, covering shape, dtype, layout, preprocessing responsibility, and output ordering. - Toolchain Handoff: Delegates quantization and compilation subtasks to the appropriate OE toolchain router (X5, S100/S100P/S600, X3, legacy) with precise inputs, baselines, and acceptance criteria, then verifies returned receipts and hashes. - Scoped Verification: Separates toolchain artifact status from sample integration status, comparing float baselines, toolchain results, and post-integration outputs on the same reproducible input. - Use Case: You trained a 3-class YOLO model and want it running in an existing 80-class sample. The Skill checks labels, output reshape, and decoding, coordinates any needed conversion, and validates the adapted sample end to end. ## Quick Start Ask the agent to integrate your custom trained model into a specific RDK Model Zoo sample, stating the target platform, model artifact path, and what changed such as class count or input shape.
