ResNet101v2

class ResNet101v2(noTop: Boolean, inputShape: IntArray?) : TFModels.CV<Functional>

This model is a neural network for image classification that take images as input and classify the major object in the image into a set of 1000 different classes (labels are available via org.jetbrains.kotlinx.dl.api.core.util.loadImageNetClassLabels method).

This model has 101 layers with ResNetv2 architecture.

The model have

an input with the shape (1x224x224x3)
an output with the shape (1x1000)

NOTE: ResNet v2 uses pre-activation function whereas ResNet v1 uses post-activation for the residual blocks.

NOTE: This model is converted from Keras.applications, the last few layers in the noTop model have been removed so that the user can fine-tune the model for his specific task.

Constructors

ResNet101v2

fun ResNet101v2(noTop: Boolean = false, inputShape: IntArray? = null)

Functions

model

open fun model(modelHub: ModelHub): Functional

Loads the model, identified by this name, from the modelHub.

preInit

open fun preInit(): InferenceModel

preprocessInput

open fun preprocessInput(imageFile: File, preprocessing: Preprocessing): FloatArray

open override fun preprocessInput(data: FloatArray, tensorShape: LongArray): FloatArray

Common preprocessing function for the Neural Networks trained on ImageNet and whose weights are available with the keras.application.

pretrainedModel

open override fun pretrainedModel(modelHub: ModelHub): ImageRecognitionModel

Returns the specially prepared pre-trained model of the type U.

Properties

channelsFirst

open override val channelsFirst: Boolean

If true it means that the second dimension is related to number of channels in image has short notation as NCWH, otherwise, channels are at the last position and has a short notation as NHWC.

inputShape

var inputShape: IntArray?

modelRelativePath

open override val modelRelativePath: String

Relative path to model for local and S3 buckets storages.

ResNet101v2

See also

Constructors

Functions

Properties