KR102305981B1 - Method for Training to Compress Neural Network and Method for Using Compressed Neural Network - Google Patents

Abstract

Disclosed are a neural network compression training method and a method using a compressed neural network.
According to one aspect of the present invention, there is provided a method of compressing a neural network, in particular, a neural network compression method of reducing the number of parameters to be learned by the neural network, thereby reducing the computational throughput and the computational speed of the neural network.

Description

{Method for Training to Compress Neural Network and Method for Using Compressed Neural Network}

Embodiments of the present invention relate to a method of compressing a neural network, in particular, a neural network compression method for reducing computational throughput and computation speed of a neural network by reducing the number of parameters to be learned by the neural network.

The content described in this section merely provides background information on the present invention and does not constitute the prior art.

Recently, an image classification method or image style transfer method based on deep learning uses a neural network having numerous channels. Such a neural network needs to learn a large number of parameters to perform a general image classification task or image style conversion method. As the number of parameters that the neural network needs to learn increases, there is a problem in that the processing throughput and power consumption of the processor increase exponentially. Therefore, it is necessary to compress the neural network in terms of processing throughput, processing speed, and power consumption of the processor.

A channel pruning method has been devised to compress neural networks. The channel pruning method is a method of training a neural network to remove small parameters (or weights) from a neural network after learning parameters until the loss function of the neural network is reduced to a specific criterion.

In particular, the conventional channel pruning method used in a convolutional neural network is performed in two steps. The first step is a step of compressing the neural network by removing a parameter with a small size among filter parameters from the trained neural network, and the second step is a step of re-training the filter parameters of the compressed neural network.

However, the existing channel pruning method considers only the size of the filter parameter without considering the size of an element included in the input feature map input to the convolutional neural network, and removes a parameter with a small size. In the convolutional neural network, even if the size of the filter parameter is small, when the size of an element included in the input feature map is large, the filter may generate a feature map including the large size element. As such, if all small filter parameters are removed regardless of the size of the input feature map, the large output feature map that can be generated by the small filter parameter is also removed, thereby degrading the performance of the neural network.

In addition, in the conventional channel pruning method, after performing the step of compressing the neural network, the step of re-training the filter parameters of the compressed neural network should be additionally performed. This is a process for compensating for performance degradation caused by the step of removing the filter parameter. Therefore, the conventional channel pruning method has a problem in that the time required to train the neural network increases because it has to go through a process of compressing and retraining the neural network.

In the embodiments of the present invention, the number of parameters for the neural network to learn by removing the filter that generates a zero feature map based on the feature map generated by the filter regardless of the size of the input feature map input to the filter in the neural network The main purpose is to reduce

Another embodiment of the present invention maintains the performance of the neural network by removing the filter generating the zero feature map after training the filter to output all zero feature maps or all non-zero feature maps for an arbitrary image. At the same time, it aims to reduce the number of learning parameters.

According to an aspect of the present invention, in a training method for compressing a neural network that generates feature maps using a plurality of filters for a plurality of batches including a plurality of images, obtaining the plurality of batches, providing an arrangement to the neural network; and updating the parameters of the neural network for each batch, wherein the updating includes: acquiring a plurality of feature maps corresponding to each image included in the batch using the neural network; classifying the plurality of feature maps corresponding to the respective images into a zero feature map and a non-zero feature map; and modifying the parameters of the neural network using the number of zero feature maps and the number of filters that output both zero feature maps or non-zero feature maps for two different images among the plurality of images, The zero feature map is a feature map in which all elements are 0, and the non-zero feature map provides a training method in which at least one of the elements is a non-zero feature map.

According to another aspect of this embodiment, there is provided an image processing method, comprising: acquiring an input image; providing the input image to a first neural network, wherein the first neural network generates feature maps for the input image using a plurality of filters; and obtaining, by a second neural network, a result of processing the image using the feature maps received from the first neural network, wherein the first neural network performs the first processing for a plurality of batches including a plurality of images. It is trained together with two neural networks, and the training method includes: obtaining a plurality of feature maps corresponding to each image included in the batch by using the first neural network for each batch; classifying the plurality of feature maps corresponding to each image into a zero feature map and a non-zero feature map; and modifying parameters of the first neural network using the number of zero feature maps and the number of filters that output both zero feature maps or non-zero feature maps for two different images among the plurality of images. However, the zero feature map is a feature map in which all elements are 0, and the non-zero feature map provides an image processing method in which at least one of the elements is a non-zero feature map.

As described above, according to an embodiment of the present invention, a filter for generating a zero feature map based on the feature map generated by the filter is removed regardless of the size of the input feature map input to the filter in the neural network, thereby making the neural network By reducing the number of parameters to learn, the computational throughput and computational speed of the neural network can be reduced.

According to another embodiment of the present invention, the performance of the neural network is improved by removing the filter generating the zero feature map after training the filter to output all zero feature maps or all non-zero feature maps for an arbitrary image. At the same time, it is possible to reduce the number of learning parameters while reducing the computational throughput and computational speed of the neural network.

1A and 1B are diagrams for explaining a channel pruning method among methods of compressing a neural network in a convolutional neural network.
2 is a diagram for explaining a process of compressing a neural network according to an embodiment of the present invention.
3 is a diagram for explaining a neural network compression method according to an embodiment of the present invention by an equation.
4 is a flowchart illustrating a compression training method of a neural network according to an embodiment of the present invention.
5 is a flowchart illustrating a method of using a compressed neural network according to an embodiment of the present invention.

Hereinafter, some embodiments of the present invention will be described in detail with reference to exemplary drawings. In adding reference numerals to the components of each drawing, it should be noted that the same components are given the same reference numerals as much as possible even though they are indicated on different drawings. In addition, in describing the present invention, if it is determined that a detailed description of a related known configuration or function may obscure the gist of the present invention, the detailed description thereof will be omitted.

In addition, in describing the components of the present invention, terms such as first, second, A, B, (a), (b), etc. may be used. These terms are only for distinguishing the components from other components, and the essence, order, or order of the components are not limited by the terms. Throughout the specification, when a part 'includes' or 'includes' a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated. . In addition, terms such as '~ unit' and 'module' described in the specification mean a unit that processes at least one function or operation, which may be implemented as hardware or software or a combination of hardware and software.

1A and 1B are diagrams for explaining a channel pruning method among methods of compressing a neural network in a convolutional neural network.

Referring to FIG. 1A , an input feature map 100, a kernel, that is, a first filter 102, an intermediate feature map 104, a second filter 106, and an output feature map 108 are is exemplified

The input feature map 100 is extracted from the input image and will be described as having three channels. Three input feature maps 100 are input to five first filters 102 . Each filter generates one intermediate feature map 104 for three channels. That is, the number of intermediate feature maps 104 is determined by the number of first filters. The five intermediate feature maps 104 are input to two second filters 106 , and each filter generates two output feature maps 108 for the five intermediate feature maps 104 .

In order to apply the channel pruning method, a filter with a small parameter included in the filter is removed. When the size of parameters included in two of the five first filters 102 is small, it is determined that the two filters are filters that do not affect the performance of the neural network, and two intermediates generated by the two filters The feature map can also be determined as a feature map that does not affect the performance of the neural network.

Referring to FIG. 1B , a process of re-training the neural network after removing two filters with small parameters is shown.

The three input feature maps 100 are input to the three first filters 112 , and the three first filters 112 generate three intermediate feature maps 114 . The three intermediate feature maps 114 are input to the two second filters 116 , and the two second filters 116 generate two output feature maps 118 .

At this time, before the two filters are removed, the two intermediate feature maps generated by the two filters are also removed together with the two filters. In this case, even if the size of the parameter included in the two filters is small, the two intermediate feature maps may include the parameter having a large size. That is, by removing the two filters, even the feature map that can affect the performance of the neural network is removed, so the performance of the neural network may be degraded.

In order to prevent the performance of the neural network from being deteriorated by removing two of the five input filters, the neural network performs a process of retraining the parameters included in the three first filters 112 .

2 is a diagram for explaining a process of compressing a neural network according to an embodiment of the present invention.

Referring to FIG. 2 , a first image 210 , a second image 220 , a neural network 230 , four filters 232 , 234 , 236 , 238 , and four first feature maps for the first image (212, 214, 216, 218) and four second feature maps (222, 224, 226, 228) for the second image are shown.

The first image 210 and the second image 220 are training data of the neural network. However, the first image 210 and the second image 220 are only one embodiment, and are not limited thereto, and may be data in which a plurality of characters are expressed in a multidimensional matrix.

The neural network 230 includes four filters 232 , 234 , 236 , and 238 , but this is only one embodiment and may include a plurality of filters. The neural network 230 receives the first image 210 and the second image 220 , and generates feature maps using four filters 232 , 234 , 236 , and 238 . The first feature maps 212 , 214 , 216 , and 218 and the second feature maps 222 , 224 , 226 , and 228 include a plurality of elements for each feature map.

The first feature maps 212 , 214 , 216 , and 218 are feature maps generated by the neural network 230 for the first image 210 , and the second feature maps 222 , 224 , 226 , 228 are neural networks 230 are feature maps generated with respect to the second image 220 . Specifically, the first filter 232 generates a feature map 212 for the first image and a feature map 222 for the second image. Similarly, the second filter 234 , the third filter 236 , and the fourth filter 238 include feature maps 214 , 216 , 218 for the first image and feature maps 224 for the second image, respectively. , 226, 228).

The first feature maps 212, 214, 216, 218 and the second feature maps 222, 224, 226, 228 are a zero feature map and a non-zero feature map. is divided into Here, the zero feature map means a feature map in which the values of all elements are 0, and the non-zero feature map means a feature map in which at least one of the elements is not zero. Some of the first feature maps 212 and 214 are non-zero feature maps, and others 216 and 218 of the first feature maps are zero feature maps. Some of the second feature maps 214 and 216 are non-zero feature maps, and others 212 and 218 of the second feature maps are zero feature maps.

The parameters of the neural network 230 are the number of zero feature maps 216 , 218 , 222 , and 228 , the first image 210 and the second image 220 of the filter 238 that outputs zero feature maps. It can be modified based in part on the number.

The training method for compressing the neural network according to an embodiment of the present invention is such that the number of filters that generate both zero feature maps or all non-zero feature maps for the first image 210 and the second image 220 increases. The parameters of the neural network 230 are modified.

A training method for compressing a neural network according to another embodiment of the present invention is a parameter of the neural network 230 such that the number of zero feature maps among a plurality of feature maps generated for the first image 210 and the second image 220 is increased. to correct

The training method for compressing a neural network according to an embodiment of the present invention may update parameters of the neural network in units of each batch for a plurality of batches. Here, the batch is a unit including a plurality of images that are training data. After updating the parameters of the neural network for a plurality of batches, two different images among the plurality of images are provided to the neural network. Among a plurality of filters included in the neural network, a filter that generates a zero feature map for two different images is removed.

Specifically, the training method for compressing a neural network provides one batch of a plurality of batches to the neural network, and acquires a plurality of feature maps corresponding to each image included in one batch by using the neural network. A plurality of feature maps corresponding to each image are divided into a zero feature map and a non-zero feature map. The parameters of the neural network may be modified based on the number of filters that output zero feature maps or non-zero feature maps and the number of zero feature maps for two different images included in one batch. Thereafter, the neural network can be trained by repeating the above process for each batch. For example, when two batches are given, and each batch includes eight images, the neural network training method can update the parameters of the neural network for one batch a total of 28 times. Number 28 is calculated from C(8, 2), and C(8, 2) means a combination of two different images among eight images.

The compressed neural network can be trained through the processes described in FIG. 2, and since training and compression of the parameters of the neural network are performed at the same time, the number of parameters the neural network learns is reduced, but a separate retraining process is not required. performance is maintained.

3 is a diagram for explaining a neural network compression method according to an embodiment of the present invention by an equation.

The neural network compression training method according to an embodiment of the present invention uses a loss function including two losses. The two losses are channel loss and xor loss (exclusive or loss).

The channel loss is a parameter that increases the number of zero feature maps among a plurality of feature maps output from the convolutional layer. That is, as the channel loss decreases, the number of zero feature maps among the output feature maps increases.

The xor loss is a parameter for removing the filter that generates the zero feature map even if the input image is different. By using xor loss, we can increase the number of filters that generate both zero feature maps or all non-zero feature maps for both input images. That is, as the xor loss is smaller, the number of filters that always output zero feature maps or all non-zero feature maps for an arbitrary input image increases.

An embodiment of the present invention can reduce the number of learning parameters of the neural network while maintaining the performance of the neural network by removing the filter that always generates a zero feature map for an arbitrary image by using the two losses.

은 L번째 컨볼루션 레이어로부터 출력되는 특징맵들이고,

은 L번째 레이어로부터 출력되는 특징맵들의 개수를 나타내는 채널이다. Referring to FIG. 3 , B is a unit for updating parameters of a neural network, and a convolution layer is included in the neural network and includes a plurality of filters, but outputs a plurality of feature maps.

are the feature maps output from the L-th convolutional layer,

is a channel indicating the number of feature maps output from the L-th layer.

와 이미지 j에 대한 복수의 특징맵인

에 대해 L0-정규화(L0-normalization)를 통해 제로 특징맵과 논제로 특징맵으로 구분한다. xor 로스는 이미지 i 및 이미지 j에 대해 모두 제로 특징맵을 생성하거나 모두 논제로 특징맵을 생성하는 필터의 수를 합한 값에 부분적으로 기초하여 계산된다. 채널 로스는 이미지 i 및 이미지 j에 대해 논제로 특징맵을 생성하는 필터의 수를 모두 합한 값에 부분적으로 기초하여 계산된다.A training method according to an embodiment of the present invention provides image i and image j included in batch B to a neural network, and acquires feature maps from a convolution layer l included in the neural network. After that, a plurality of feature maps for image i

and multiple feature maps for image j.

is divided into a zero feature map and a non-zero feature map through L0-normalization. The xor loss is calculated based in part on the sum of the number of filters that generate both zero feature maps or all non-zero feature maps for image i and image j. The channel loss is calculated based in part on the sum of the number of filters that generate non-zero feature maps for image i and image j.

는 L0-norm으로서, 특징맵 x에 포함된 모든 엘리먼트가 0이면 0을 반환하고, 엘리먼트 중 적어도 하나가 0이 아니면 1을 반환하는 기호이다. 또한, 수학식 2에서 t는 채널의 인덱스(index)를 의미한다.Specifically, the channel loss and the xor loss are calculated by Equations 1 and 2, respectively. In Equation 1 and Equation 2

is a L0-norm, a symbol that returns 0 if all elements included in the feature map x are 0, and returns 1 if at least one of the elements is not 0. Also, in Equation 2, t means an index of a channel.

는 수학식 3과 같이 수정될 수 있다. 수학식 3에서

는 L2-norm(L2 정규화)이다. 이는, 채널 로스 및 xor 로스를 포함하는 손실함수를 미분가능하도록 하기 위함이다.

일 수 있다.According to one embodiment of the present invention.

can be modified as in Equation (3). in Equation 3

is L2-norm (L2 normalized). This is to make a loss function including a channel loss and an xor loss differentiable.

can be

4 is a flowchart illustrating a compression training method of a neural network according to an embodiment of the present invention.

The processes described below may be performed using one or more memories and one or more processors.

First, a plurality of images is acquired from a memory for storing the plurality of images (S400). In this case, images may be acquired in units of batches including a plurality of images.

A plurality of batches are provided to the neural network, and a plurality of feature maps generated by the neural network using a plurality of filters for each image included in one batch are received ( S402 ).

A plurality of feature maps corresponding to each image are divided into a zero feature map and a non-zero feature map (S404).

The parameters of the neural network are modified to increase the number of filters that generate all-zero feature maps or all-non-zero feature maps for two different images among the plurality of images (S406).

The parameters of the neural network are modified so that the number of zero feature maps among the plurality of feature maps is increased (S408).

The neural network training method according to an embodiment of the present invention may repeat steps S402 to S408 for each batch. Also, the neural network training method may simultaneously perform steps S406 and S408.

5 is a flowchart illustrating a method of using a compressed neural network according to an embodiment of the present invention.

As a method of processing an image using a compressed neural network, a method using a trained first neural network and a trained second neural network will be described.

Referring to FIG. 5 , a plurality of input images is acquired from a memory ( S500 ).

The plurality of acquired input images are provided to the first neural network (S502). The first neural network generates feature maps for each of a plurality of input images by using a plurality of filters. The feature maps generated by the first neural network are transmitted to the second neural network.

The second neural network obtains a result of image processing using the feature maps received from the first neural network (S504).

Here, when the second neural network is an image classifier, a result obtained by classifying a plurality of images by the second neural network may be obtained.

On the other hand, when the second neural network is an image style converter, it is possible to obtain a composite image to which image style conversion is applied to two different images among a plurality of images. Image style conversion refers to a technique of extracting content features from one image among two images, extracting style features from the other image, and outputting an image in which content features and style features are synthesized.

Meanwhile, the first neural network and the second neural network may be trained together. It will be described below.

The training method first acquires a plurality of feature maps corresponding to each image included in the batch by using a first neural network for each batch among a plurality of batches. The plurality of feature maps generated by the first neural network are transmitted to the second neural network. Thereafter, a plurality of feature maps corresponding to each image are divided into a zero feature map and a non-zero feature map. The parameters of the first neural network are modified by using the number of zero feature maps and the number of filters that output all zero feature maps or all non-zero feature maps for two different images among the plurality of images.

When the first neural network and the second neural network are trained to classify a plurality of images according to an embodiment of the present invention, the second neural network may be an image classifier. In the training method, first, a plurality of images are provided to the first neural network. The first neural network generates a plurality of feature maps for a plurality of images and transmits them to the second neural network, and the second neural network labels each of the plurality of images using the plurality of feature maps. A loss function is extracted from a plurality of labeled images and preset correct answer data. The channel loss and xor loss of the first neural network are added to the extracted loss function. Finally, the first neural network and the second neural network may be trained by modifying the parameters of the first and second neural networks together to reduce the loss function including the channel loss and the xor loss to a preset value.

On the other hand, when the first neural network and the second neural network are trained to transform an image style according to an embodiment of the present invention, the first neural network may be an encoder and the second neural network may be a decoder.

In the training method, first, two different images are provided to the first neural network. The first neural network extracts a content-related feature map from one image, and extracts a style-related feature map from the other image. The first neural network transmits a feature map related to content and a feature map related to a style to the second neural network. The second neural network generates an image in which content and style are synthesized by using a feature map related to content and a feature map related to style. A loss function is extracted from the synthesized image and preset correct answer data. The channel loss and xor loss of the first neural network are added to the extracted loss function. Finally, the first neural network and the second neural network may be trained by modifying the parameters of the first and second neural networks together to reduce the loss function including the channel loss and the xor loss to a preset value.

Although it is described that steps S400 to S504 are sequentially executed in FIGS. 4 and 5 , this is merely illustrative of the technical idea of an embodiment of the present invention. In other words, those of ordinary skill in the art to which an embodiment of the present invention pertain change the order described in FIGS. 4 and 5 within a range that does not deviate from the essential characteristics of an embodiment of the present invention, or process S400 to process S400 Since it will be possible to apply various modifications and variations to parallel execution of one or more processes in S504, FIGS. 4 and 5 are not limited to a time-series order.

Meanwhile, the processes shown in FIGS. 4 and 5 can be implemented as computer-readable codes on a computer-readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data readable by a computer system is stored. That is, the computer-readable recording medium includes a magnetic storage medium (eg, a ROM, a floppy disk, a hard disk, etc.) and an optical readable medium (eg, a CD-ROM, a DVD, etc.). In addition, the computer-readable recording medium is distributed in a network-connected computer system so that the computer-readable code can be stored and executed in a distributed manner. In addition, one or more processors and one or more memories for storing instructions, wherein when the instructions are executed by the one or more processors, the instructions cause the one or more processors to perform steps S400 to S408 or S500 to S504. to do it

The above description is merely illustrative of the technical idea of this embodiment, and various modifications and variations will be possible by those skilled in the art to which this embodiment belongs without departing from the essential characteristics of the present embodiment. Accordingly, the present embodiments are intended to explain rather than limit the technical spirit of the present embodiment, and the scope of the technical spirit of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the following claims, and all technical ideas within the equivalent range should be interpreted as being included in the scope of the present embodiment.

100: input feature map 108: output feature map
210: first image 220: second image
230: neural network

Claims (16)

In a training method for compressing a neural network that generates feature maps using a plurality of filters for a plurality of batches including a plurality of images,
obtaining the plurality of batches and providing the plurality of batches to the neural network; and
It includes the process of updating the parameters of the neural network for each batch,
The update process is
obtaining a plurality of feature maps corresponding to each image included in the arrangement by using the neural network;
classifying the plurality of feature maps corresponding to the respective images into a zero feature map and a non-zero feature map; and
A process of modifying the parameters of the neural network using the number of filters that output all zero feature maps or all non-zero feature maps for two different inputs, and the number of zero feature maps,
The zero feature map is a feature map in which all elements are 0, and the non-zero feature map is a feature map in which at least one of the elements is non-zero. According to claim 1,
The neural network includes at least one convolution layer, and the plurality of filters are included in the at least one convolution layer. According to claim 1,
The modification process is
and modifying the parameter to increase the number of zero feature maps and the number of filters. An image processing method comprising:
acquiring an input image;
providing the input image to a first neural network, wherein the first neural network generates feature maps for the input image using a plurality of filters; and
a second neural network obtaining a result of processing the image using the feature maps received from the first neural network,
The first neural network is trained together with the second neural network for a plurality of batches including a plurality of images, and the training method is for each batch.
obtaining a plurality of feature maps corresponding to each image included in the arrangement by using the first neural network;
classifying the plurality of feature maps corresponding to each image into a zero feature map and a non-zero feature map; and
modifying parameters of the first neural network using the number of filters that output all zero feature maps or all non-zero feature maps for two different inputs, and the number of zero feature maps;
including,
The zero feature map is a feature map in which all elements are 0, and the non-zero feature map is a feature map in which at least one of the elements is non-zero. 5. The method of claim 4,
The first neural network includes at least one convolutional layer, and the plurality of filters are included in the at least one convolutional layer. 5. The method of claim 4,
The modification process is
and modifying the parameter to increase the number of the zero feature maps and the number of filters. 5. The method of claim 4,
The first neural network is an encoder, and the second neural network is a decoder. 5. The method of claim 4,
The first neural network is an encoder, and the second neural network is an image classifier. A system for compressing a neural network, comprising:
one or more processors; and
one or more memories for storing instructions;
when the instructions are executed by the one or more processors, the instructions cause the one or more processors to perform a neural network training method;
The training method is
A training method for compressing the neural network that generates feature maps using a plurality of filters for a plurality of batches including a plurality of images,
obtaining the plurality of batches and providing the plurality of batches to the neural network; and
It includes the process of updating the parameters of the neural network for each batch,
The update process is
obtaining a plurality of feature maps corresponding to each image included in the arrangement by using the neural network;
classifying the plurality of feature maps corresponding to the respective images into a zero feature map and a non-zero feature map; and
A process of modifying the parameters of the neural network using the number of filters that output all zero feature maps or all non-zero feature maps for two different inputs, and the number of zero feature maps,
The zero feature map is a feature map in which all elements are 0, and the non-zero feature map is a feature map in which at least one of the elements is non-zero. 10. The method of claim 9,
wherein the neural network includes at least one convolutional layer, and the plurality of filters are included in the at least one convolutional layer. 10. The method of claim 9,
The modification process is
and modifying the parameter to increase the number of zero feature maps and the number of filters. A system for processing an image, comprising:
one or more processors; and
one or more memories for storing instructions;
when the instructions are executed by the one or more processors, the instructions cause the one or more processors to perform an image processing method,
The image processing method is
acquiring an input image;
providing the input image to a first neural network, wherein the first neural network generates feature maps for the input image using a plurality of filters; and
a second neural network obtaining a result of processing the image using the feature maps received from the first neural network,
The first neural network is trained together with the second neural network for a plurality of batches including a plurality of images, and the training method is for each batch.
obtaining a plurality of feature maps corresponding to each image included in the arrangement by using the first neural network;
classifying the plurality of feature maps corresponding to each image into a zero feature map and a non-zero feature map; and
modifying parameters of the first neural network using the number of filters that output all zero feature maps or all non-zero feature maps for two different inputs, and the number of zero feature maps;
includes,
The zero feature map is a feature map in which all elements are 0, and the non-zero feature map is a feature map in which at least one of the elements is non-zero. 13. The method of claim 12,
The first neural network includes at least one convolutional layer, and the plurality of filters are included in the at least one convolutional layer. 13. The method of claim 12,
The modification process is
and modifying the parameter to increase the number of zero feature maps and the number of filters. 13. The method of claim 12,
The first neural network is an encoder, and the second neural network is a decoder. 13. The method of claim 12,
The first neural network is an encoder, and the second neural network is an image classifier.

Images