ai.djl.basicmodelzoo.cv.classification

Class NiN

java.lang.Object

ai.djl.basicmodelzoo.cv.classification.NiN

public final class NiN
extends java.lang.Object

NiN uses convolutional layers with window shapes of 11×11 , 5×5 , and 3×3 , and the corresponding numbers of output channels are the same as in AlexNet. Each NiN block is followed by a maximum pooling layer with a stride of 2 and a window shape of 3×3 .

The conventional convolutional layer uses linear filters followed by a nonlinear activation function to scan the input.

NiN model from the "Network In Network" http://arxiv.org/abs/1312.4400 paper.

See Also:

The D2L chapter on NiN

Nested Class Summary

Nested Classes

Modifier and Type	Class and Description
static class	NiN.Builder The Builder to construct a NiN object.

Method Summary

Modifier and Type	Method and Description
static NiN.Builder	builder() Creates a builder to build a NiN.
static ai.djl.nn.Block	niN(NiN.Builder builder) The NiN block consists of one convolutional layer followed by two 1×1 convolutional layers that act as per-pixel fully-connected layers with ReLU activations.
ai.djl.nn.SequentialBlock	niNBlock(int numChannels, ai.djl.ndarray.types.Shape kernelShape, ai.djl.ndarray.types.Shape strideShape, ai.djl.ndarray.types.Shape paddingShape) Creates a constituent NiN block that becomes a part of the whole NiN model.

Methods inherited from class java.lang.Object

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

Method Detail

niN

public static ai.djl.nn.Block niN(NiN.Builder builder)

The NiN block consists of one convolutional layer followed by two 1×1 convolutional layers that act as per-pixel fully-connected layers with ReLU activations. The convolution width of the first layer is typically set by the user. The subsequent widths are fixed to 1×1.

Parameters:

builder - the NiN.Builder with the necessary arguments.

Returns:

a NiN block.

builder

public static NiN.Builder builder()

Creates a builder to build a NiN.

Returns:

a new builder

niNBlock

public ai.djl.nn.SequentialBlock niNBlock(int numChannels,
                                          ai.djl.ndarray.types.Shape kernelShape,
                                          ai.djl.ndarray.types.Shape strideShape,
                                          ai.djl.ndarray.types.Shape paddingShape)

Creates a constituent NiN block that becomes a part of the whole NiN model.

Parameters:

numChannels - the number of channels in a NiN block.

kernelShape - kernel Shape in the 1st convolutional layer of a NiN block.

strideShape - stride Shape in a NiN block.

paddingShape - padding Shape in a NiN block.

Returns:

a constituent niN block.