Tag - Model Quantization
ESP32-S3 can run TinyML, but production success depends less on AI instructions alone and more on SRAM, tensor arena sizing, INT8 quantization, operator support, PSRAM latency, sensor pipelines, and real-time inference budgets. This article explains the practical bottlenecks and boundaries.
A step-by-step guide to convert MediaPipe Gesture Recognition models to RKNN and running real-time inference on RK3566 NPU, with code and troubleshooting tips.
AI and Machine Learning, Blog , December 2, 2025 , Computer Vision, Edge AI, embedded AI, Gesture Recognition, Landmark Detection, MediaPipe, Model Quantization, RK3566, RKNN, Rockchip