Class RetinaNet

Represents a RetinaNet model. Rather than creating directly, it is intended to be created using load().

Hierarchy

RetinaNet

Index

Constructors

constructor

Properties

Methods

Constructors

constructor

new RetinaNet(model: LayersModel, classes: string[], preprocessingMode: string, anchorParams?: AnchorParameters): RetinaNet

- Defined in lib/retinanet.ts:70
Parameters
- model: LayersModel
- classes: string[]
- preprocessingMode: string
- Default value anchorParams: AnchorParameters = defaultAnchorParameters
Returns RetinaNet

Properties

Protected anchorParams

anchorParams: AnchorParameters

Protected classes

classes: string[]

Protected height

height: number

model

model: LayersModel

Protected preprocessingMode

preprocessingMode: string

Protected width

width: number

Methods

detect

detect(img: tf.Tensor3D | ImageData | HTMLImageElement | HTMLCanvasElement | HTMLVideoElement, threshold?: number, nmsThreshold?: number): Promise<DetectedObject[]>

- Defined in lib/retinanet.ts:103
Computes predictions. We currently do not support class-specific filtering. When non-max suppression is applied, it will be across all boxes, regardless of class.

Parameters
- img: tf.Tensor3D | ImageData | HTMLImageElement | HTMLCanvasElement | HTMLVideoElement
  
  The image object on which to run object detection
- Default value threshold: number = 0.5
  
  The prediction threshold
- Default value nmsThreshold: number = 0.5
  
  The non-max suppresion IoU threshold
Returns Promise<DetectedObject[]>

dispose

dispose(): void

- Defined in lib/retinanet.ts:199
Remove the model from memory.

Returns void

Private handleImageTensor

handleImageTensor(imageTensor: tf.Tensor3D): [tf.Tensor4D, number, number]

- Defined in lib/retinanet.ts:203
Parameters
- imageTensor: tf.Tensor3D
Returns [tf.Tensor4D, number, number]

Generated using TypeDoc