Custom Algorithm
A full-workflow tutorial and toolchain for training vision, hearing and smell AI algorithms and deploying them, as packaged algorithm modules, onto a specific edge AI platform.
Custom Algorithm provides multimodal algorithm training source code and an algorithm toolchain for edge AI devices, covering three perception directions — vision (object detection, image classification, instance segmentation, object tracking), hearing (sound event recognition) and smell (sensor data analysis) — aimed at applications such as camera video analysis, industrial supervision, park security and abnormal-sound recognition.
It targets the complete workflow from algorithm training to engineering deployment: data preparation, model training, model conversion, algorithm packaging, importing the package onto an edge device, inference scheduling, alarm linkage and on-site operation, rather than only getting a model to run in a lab. It supports Zhiquli's ks968 edge AI platform, a customized board based on a Rockchip SoC, and aims to help a developer combine a self-trained model with a complete software and hardware system for real commercial deployment.
It is written in Python.