ENOT User Manual
Embedded Network Optimization Technology
ENOT, or Embedded Network Optimization Technology, is a flexible tool for Deep Learning developers which automates neural architecture optimization. It can be useful in the following scenarios:
Target metric maximization (e.g., classification accuracy or intersection over union);
Target metric maximization with constrained computational resources (e.g., RAM, latency);
Pareto frontier (latency vs target metric) search.
Framework advantages:
Controlled ratio between latency and network performance;
Networks in the pre-trained search space can exceed their stand-alone variants (in some scenarios);
Compatibility with almost any DL task and simple integration with the existing training pipelines.
Before reading the documentation, we recommend you to look at the Tutorials to clarify basic notions and concepts of the framework.
Other ENOT LLC products:
enot-lite for easy-to-use inference on CPUs and NVIDIA GPUs.
enot_yolov5 - ENOT framework integration into YOLOv5 for Neural Architecture Search of YOLOv5-like models.
mmdetection-enot - ENOT framework integration into MMDetection for Neural Architecture Search.