Overview

A responsive sketch recognition demo built on a compact industrial platform.
Users draw directly on the touchscreen, and the system recognises sketches in real time from 345 categories, all processed locally on the device.
The demo runs on the RZ/V2N SOM using the integrated DRP-AI3 accelerator, with a MobileNetV2 classifier quantised to INT8 for efficient performance and low power consumption. No external GPU or cloud processing is required. Everything runs directly on the board.

Why It Matters

Real-Time Touchscreen Recognition: The platform provides fast recognition with minimal delay between drawing and prediction, creating a smooth and interactive user experience.

Fully Local Processing: All processing stays on-device without relying on cloud connectivity, making the platform suitable for applications where responsiveness and local operation are important.

Compact and Cost-Effective Platform: The RZ/V2N with DRP-AI3 delivers efficient inference performance in a compact industrial form factor without requiring a discrete GPU or external accelerator card. This makes it well suited for scalable embedded and industrial deployments.

Developer-Friendly Workflow: The demo includes a complete workflow covering:
Train → Quantise → Compile → Cross-Compile → Deploy → Run
The process is fully documented and can be used as a reference for additional embedded vision and recognition applications.

Highlight

Value

AI engine

DRP-AI3 — 15 Sparse TOPS / 4 Dense TOPS

CPU

4× Arm Cortex-A55 @ up to 1.8 GHz + 1× Cortex-M33 (real-time)

Memory

Up to 8 GB LPDDR4 with inline ECC

Form factor

47 × 30 mm, industrial −40 °C to +85 °C

Highlight

Value

Industrial I/O

RS232/RS485 · CAN-FD · 2× GbE · USB 3.2 · PoE 802.3at

Display + camera

MIPI-DSI · MIPI-CSI 4-lane

Power

7–32 V wide-input, reverse-polarity protected, or PoE

Operating + OS

Industrial −40 °C to +85 °C · Yocto Linux + Weston

Property

Value

Architecture

MobileNetV2 — ImageNet-pretrained backbone, custom 1280 → 768 → 345 head

Dataset

Google Quick, Draw! — 345 classes

Input

[1, 3, 128, 128] (NCHW), grayscale replicated to 3 channels

Quantisation

INT8 — Percentile 99.99 calibration over 1,725 representative PNGs

Compiler

DRP-AI Translator i8 v1.11 → DRP-AI TVM v2.7+ (MERA2 backend)

Inference

~1 ms on DRP-AI3

Training

Two-stage transfer learning — ~6 hours on RTX 5060 Ti — ~82 % validation accuracy

Developer Access

Our developers have made the full implementation details, setup instructions, and technical documentation for this demo available here:
https://dev.solid-run.com/renesas/rz-v2n/rz-v2n-other-articles/ai-demo-quick-draw-doodle-recognition-rz-v2n
This page includes the complete workflow for running the demo on the RZ/V2N platform, along with deployment steps and system configuration details.