Model Zoo

This section describes the AI model packages available for the RZ/V2H RDK platform.

These packages provide a unified C++ framework for deploying AI models optimized with DRP-AI acceleration, enabling high power-efficiency inference on the RZ/V2H architecture.

Overview

The RZ/V2H RDK AI model ecosystem is organized into the following layers:

Base Framework (rzv_model): Core C++ library providing the BaseModel class, DRP-AI runtime integration, and shared utilities for model loading, preprocessing, inference, and postprocessing.
Model-Specific Packages: Each AI model family (YOLOX - rzv_yolox, YOLOv8 - rzv_yolov8, HRNetV2 - rzv_hrnetv2, RTMPose - rzv_rtmpose, MediaPipe - rzv_mediapipe, Gold-YOLO - rzv_gold_yolo) is implemented as a separate package that extends the base framework with task-specific postprocessing logic.
ROS 2 Application Packages (rzv_object_detection, rzv_pose_estimation): ROS 2 nodes that combine the model packages with camera/image input and visualization output.
ROS 2 Utilities (rzv_model_utils_ros2): Helper functions for integrating AI models into ROS 2 applications, including model configuration loading, message encoding, and diagnostics.

Available Packages

Core Packages

Package	Description	License
rzv_model	Base framework providing `BaseModel` class, DRP-AI runtime integration, and shared utilities for model loading, preprocessing, inference, and postprocessing.	Apache 2.0
rzv_model_utils_ros2	ROS 2 utility library for model configuration loading, message encoding, and inference diagnostics.	Apache 2.0

Object Detection Model Packages

Package	Description	License
rzv_yolox	YOLOX object detection with axis-aligned bounding boxes. Extends `BaseModel` with YOLOX-specific postprocessing.	Apache 2.0
rzv_yolov8	YOLOv8 detection (axis-aligned and oriented bounding boxes) with DFL decoding and multi-threaded postprocessing.	AGPLv3
rzv_gold_yolo	Gold-YOLO object detection with axis-aligned bounding boxes.	Apache 2.0

Pose Estimation Model Packages

Package	Description	License
rzv_hrnetv2	HRNetV2-based pose estimation with heatmap decoding for keypoint detection.	Apache 2.0
rzv_rtmpose	RTMPose lightweight real-time pose estimation.	Apache 2.0
rzv_mediapipe	MediaPipe hand landmark detection with handedness classification.	Apache 2.0

ROS 2 Application Packages

The following packages can be used as examples of how to integrate model packages into ROS 2 applications. You can also refer to the source code of these packages for implementation details and best practices.

Package	Description	License
rzv_object_detection	ROS 2 nodes for camera-based and static image object detection using YOLOX, YOLOv8, and Gold-YOLO models.	Apache 2.0
rzv_pose_estimation	ROS 2 nodes for hand landmark estimation with a two-stage pipeline (detection + landmark).	Apache 2.0

Model Zoo Contents