org.deeplearning4j.nn.conf.layers.objdetect

Class Yolo2OutputLayer

java.lang.Object

org.deeplearning4j.nn.conf.layers.Layer

org.deeplearning4j.nn.conf.layers.objdetect.Yolo2OutputLayer

All Implemented Interfaces:

Serializable, Cloneable, TrainingConfig

public class Yolo2OutputLayer
extends Layer

Output (loss) layer for YOLOv2 object detection model, based on the papers: YOLO9000: Better, Faster, Stronger - Redmon & Farhadi (2016) - https://arxiv.org/abs/1612.08242
and
You Only Look Once: Unified, Real-Time Object Detection - Redmon et al. (2016) - http://www.cv-foundation.org/openaccess/content_cvpr_2016/papers/Redmon_You_Only_Look_CVPR_2016_paper.pdf
This loss function implementation is based on the YOLOv2 version of the paper. However, note that it doesn't currently support simultaneous training on both detection and classification datasets as described in the YOlO9000 paper.
Note: Input activations to the Yolo2OutputLayer should have shape: [minibatch, b*(5+c), H, W], where:
b = number of bounding boxes (determined by config - see papers for details)
c = number of classes
H = output/label height
W = output/label width

Important: In practice, this means that the last convolutional layer before your Yolo2OutputLayer should have output depth of b*(5+c). Thus if you change the number of bounding boxes, or change the number of object classes, the number of channels (nOut of the last convolution layer) needs to also change.
Label format: [minibatch, 4+C, H, W]
Order for labels depth: [x1,y1,x2,y2,(class labels)]
x1 = box top left position
y1 = as above, y axis
x2 = box bottom right position
y2 = as above y axis
Note: labels are represented as a multiple of grid size - for a 13x13 grid, (0,0) is top left, (13,13) is bottom right
Note also that mask arrays are not required - this implementation infers the presence or absence of objects in each grid cell from the class labels (which should be 1-hot if an object is present, or all 0s otherwise).

Author:

Alex Black

See Also:

Serialized Form

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	Yolo2OutputLayer.Builder

Field Summary

Fields inherited from class org.deeplearning4j.nn.conf.layers.Layer

constraints, iDropout, layerName

Method Summary

Modifier and Type	Method and Description
GradientNormalization	getGradientNormalization()
double	getGradientNormalizationThreshold()
LayerMemoryReport	getMemoryReport(InputType inputType) This is a report of the estimated memory consumption for the given layer
InputType	getOutputType(int layerIndex, InputType inputType) For a given type of input to this layer, what is the type of the output?
InputPreProcessor	getPreProcessorForInputType(InputType inputType) For the given type of input to this layer, what preprocessor (if any) is required? Returns null if no preprocessor is required, otherwise returns an appropriate InputPreProcessor for this layer, such as a CnnToFeedForwardPreProcessor
List<Regularization>	getRegularizationByParam(String paramName) Get the regularization types (l1/l2/weight decay) for the given parameter.
ParamInitializer	initializer()
Layer	instantiate(NeuralNetConfiguration conf, Collection<TrainingListener> trainingListeners, int layerIndex, INDArray layerParamsView, boolean initializeParams, org.nd4j.linalg.api.buffer.DataType networkDataType)
boolean	isPretrainParam(String paramName) Is the specified parameter a layerwise pretraining only parameter? For example, visible bias params in an autoencoder (or, decoder params in a variational autoencoder) aren't used during supervised backprop. Layers (like DenseLayer, etc) with no pretrainable parameters will return false for all (valid) inputs.
void	setNIn(InputType inputType, boolean override) Set the nIn value (number of inputs, or input channels for CNNs) based on the given input type

Methods inherited from class org.deeplearning4j.nn.conf.layers.Layer

clone, getUpdaterByParam, initializeConstraints, resetLayerDefaultConfig, setDataType

Methods inherited from class java.lang.Object

equals, finalize, getClass, hashCode, notify, notifyAll, toString, wait, wait, wait

Methods inherited from interface org.deeplearning4j.nn.api.TrainingConfig

getLayerName

Method Detail

instantiate

public Layer instantiate(NeuralNetConfiguration conf,
                         Collection<TrainingListener> trainingListeners,
                         int layerIndex,
                         INDArray layerParamsView,
                         boolean initializeParams,
                         org.nd4j.linalg.api.buffer.DataType networkDataType)

Specified by:

instantiate in class Layer

initializer

public ParamInitializer initializer()

Specified by:

initializer in class Layer

Returns:

The parameter initializer for this model

getOutputType

public InputType getOutputType(int layerIndex,
                               InputType inputType)

Description copied from class: Layer

For a given type of input to this layer, what is the type of the output?

Specified by:

getOutputType in class Layer

Parameters:

layerIndex - Index of the layer

inputType - Type of input for the layer

Returns:

Type of output from the layer

setNIn

public void setNIn(InputType inputType,
                   boolean override)

Description copied from class: Layer

Set the nIn value (number of inputs, or input channels for CNNs) based on the given input type

Specified by:

setNIn in class Layer

Parameters:

inputType - Input type for this layer

override - If false: only set the nIn value if it's not already set. If true: set it regardless of whether it's already set or not.

getPreProcessorForInputType

public InputPreProcessor getPreProcessorForInputType(InputType inputType)

Description copied from class: Layer

For the given type of input to this layer, what preprocessor (if any) is required?
Returns null if no preprocessor is required, otherwise returns an appropriate InputPreProcessor for this layer, such as a CnnToFeedForwardPreProcessor

Specified by:

getPreProcessorForInputType in class Layer

Parameters:

inputType - InputType to this layer

Returns:

Null if no preprocessor is required, otherwise the type of preprocessor necessary for this layer/input combination

getRegularizationByParam

public List<Regularization> getRegularizationByParam(String paramName)

Description copied from class: Layer

Get the regularization types (l1/l2/weight decay) for the given parameter. Different parameters may have different regularization types.

Specified by:

getRegularizationByParam in interface TrainingConfig

Specified by:

getRegularizationByParam in class Layer

Parameters:

paramName - Parameter name ("W", "b" etc)

Returns:

Regularization types (if any) for the specified parameter

isPretrainParam

public boolean isPretrainParam(String paramName)

Description copied from class: Layer

Is the specified parameter a layerwise pretraining only parameter?
For example, visible bias params in an autoencoder (or, decoder params in a variational autoencoder) aren't used during supervised backprop.
Layers (like DenseLayer, etc) with no pretrainable parameters will return false for all (valid) inputs.

Specified by:

isPretrainParam in interface TrainingConfig

Specified by:

isPretrainParam in class Layer

Parameters:

paramName - Parameter name/key

Returns:

True if the parameter is for layerwise pretraining only, false otherwise

getGradientNormalization

public GradientNormalization getGradientNormalization()

Returns:

The gradient normalization configuration

getGradientNormalizationThreshold

public double getGradientNormalizationThreshold()

Returns:

The gradient normalization threshold

getMemoryReport

public LayerMemoryReport getMemoryReport(InputType inputType)

Description copied from class: Layer

This is a report of the estimated memory consumption for the given layer

Specified by:

getMemoryReport in class Layer

Parameters:

inputType - Input type to the layer. Memory consumption is often a function of the input type

Returns:

Memory report for the layer