Models

Models allow deep learning analytics to be performed on individual video frame data.

Infer Element

The infer element provides access to Nimble's deep learning backend.

name: infer
args:
  model: model-name
  hw: cpu
  mode: a
  score_threshold: 0.3
  iou_threshold: 0.3

Hardware Support

Nimble provides first-class support for three different hardware targets: cpu, igpu and gpu, and second-level support for asic. Each model lists the hardware targets they support, please contact a Megh representative if you are interested in asic devices. Specify auto to allow Nimble to automatically load balance between cpu and igpu accelerators. This is the default.

Queueing Mode

Models can be configured to operate in one of two modes: asynchronous (a) or synchronous (s). Models operating in asynchronous mode are generally used for high-throughput workloads whereas synchronous mode offers lower latency for those workloads that are sensitive to inference time.

Confidence Thresholding

Two thresholding parameters are available for filtering the inference results before they are added to the pipeline metadata: score_threshold and iou_threshold.

Built-in Models

A good starting point for person detection are the Megh Models. They have been trained on our internal datasets and are optimised for surveillance camera and eye level deployments in street, factory, and supermarket scenarios. Each model has different SKUs and can be tailored to your accuracy and hardware requirements.