WO2019076109A1 - 图像信息的池化方法及装置、存储介质、处理器 - Google Patents

图像信息的池化方法及装置、存储介质、处理器 Download PDF

Info

Publication number
WO2019076109A1
WO2019076109A1 PCT/CN2018/099597 CN2018099597W WO2019076109A1 WO 2019076109 A1 WO2019076109 A1 WO 2019076109A1 CN 2018099597 W CN2018099597 W CN 2018099597W WO 2019076109 A1 WO2019076109 A1 WO 2019076109A1
Authority
WO
WIPO (PCT)
Prior art keywords
pooling
processed
image information
image
result
Prior art date
Application number
PCT/CN2018/099597
Other languages
English (en)
French (fr)
Inventor
高灵波
黄钦
刘文峰
陈恒
Original Assignee
格力电器（武汉）有限公司
珠海格力电器股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 格力电器（武汉）有限公司, 珠海格力电器股份有限公司 filed Critical 格力电器（武汉）有限公司
Publication of WO2019076109A1 publication Critical patent/WO2019076109A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T1/00—General purpose image data processing
    • G06T1/60—Memory management
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/04—Context-preserving transformations, e.g. by using an importance map

Definitions

  • the present application relates to the field of image processing, and in particular to a method and apparatus for pooling image information, a storage medium, and a processor.
  • the embodiment of the present application provides a pooling method and device for storing image information, a storage medium, and a processor, so as to at least solve the problem that since all data waiting for the convolution layer is completed, the data is input to the pooling layer for pooling processing.
  • the technical problem caused by the need to occupy a large read bandwidth.
  • a method for pooling image information includes: acquiring at least one image to be processed in a pooled window; performing first pooling processing on the at least one image to be processed At least one pooling result; buffering the at least one pooling result to the buffer; performing second pooling processing on the at least one previous pooling result in the buffer and the at least one pooling result, to obtain the pooling window Pooled results.
  • the method before performing the second pooling process on the at least one previous pooling result of the buffer and the at least one pooling result to obtain the pooling result of the pooling window, the method further includes: determining the buffer Whether the number of cached pooling results is greater than or equal to a preset value; if the number of pooled results buffered in the buffer is greater than the preset value, the second pooling process is performed.
  • the method further includes: detecting whether the at least one image information to be processed completes a convolution operation; and if the at least one image information to be processed completes the convolution operation, executing Performing a first pooling process on the at least one image to be processed; if there is image information to be processed in which the convolution operation is not completed in the at least one image to be processed, the convolution has been completed in the at least one image to be processed.
  • the processed image to be processed is subjected to the first pooling process to obtain a partial pooling result; the partial pooling result is buffered to the buffer; when the new image information to be processed in the pooling window is subjected to a convolution operation, The partial pooling result stored in the buffer and the new image information to be processed are subjected to the second pooling process.
  • the method before the acquiring at least one image information to be processed in the pooling window, the method further includes: performing convolution operation on the at least one image information to be processed to obtain a convolution operation result; and storing the convolution operation result.
  • the method before performing the convolution operation on the at least one image information to be processed, the method further includes: acquiring an original image; converting the original image into at least one image to be processed; acquiring according to the size of the at least one image to be processed. At least one of the image information to be processed.
  • the acquiring the at least one image to be processed according to the size of the at least one image to be processed includes: if the size of the image to be processed is smaller than a preset value, extracting contour information from the image to be processed as the to-be-processed The image information is processed; if the size of the image to be processed is greater than or equal to the preset value, the detail information is extracted from the image to be processed as the image information to be processed.
  • a pooling apparatus for image information including: an acquiring unit, configured to acquire at least one image information to be processed in a pooling window; and a processing unit, configured to use the at least one The image information to be processed is subjected to the first pooling process to obtain at least one pooling result; the buffer unit is configured to cache the at least one pooling result to the buffer; and the processing unit is further configured to: at least one of the buffers The previous pooling result and the at least one pooling result are subjected to the second pooling process to obtain the pooling result of the pooling window.
  • a pooling device for image information, comprising: a pooling window; a processor, wherein the processor runs a program, wherein the program runs for output from the pooling window
  • the data is executed as follows: acquiring at least one image to be processed in the pooling window; performing a first pooling process on the at least one image to be processed to obtain at least one pooling result; and caching the at least one pooling result And to the buffer; performing a second pooling process on the at least one previous pooling result in the buffer and the at least one pooling result to obtain a pooling result of the pooling window.
  • a storage medium comprising a stored program, wherein the program executes a pooling method of image information having any of the above features.
  • a processor for running a program wherein the program is executed to perform a pooling method of image information having any of the above features.
  • At least one image to be processed in the pooling window is acquired; performing a first pooling process on the at least one image to be processed to obtain at least one pooling result; and buffering at least one pooling result into the buffer And performing a second pooling process on the at least one previous pooling result in the buffer and the at least one pooling result, and obtaining the pooling result of the pooling window by first processing the first pooled at least one pool
  • the result is cached to the buffer, and then all the pooled results stored in the buffer are subjected to the second pooling operation to obtain the final pooling result, so that the pool can be performed without waiting for the convolution layer to complete all the operations.
  • the purpose of the processing is to achieve the technical effect of reducing the read bandwidth, and further solve the problem that the input of the pooled layer to the pooling process is required to be occupied because all the data of the convolution layer needs to be completed.
  • the technical problem of reading bandwidth is to achieve the technical effect of reducing the read bandwidth, and further solve the problem that the input of the pooled layer to the pooling process is required to be occupied because all the data of the convolution layer needs to be completed.
  • FIG. 1 is a schematic flowchart diagram of an optional pooling method for image information according to an embodiment of the present application
  • FIG. 2 is a schematic structural diagram of an optional pooling device for image information according to an embodiment of the present application.
  • a method embodiment of a method for pooling image information is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings may be executed in a computer system such as a set of computer executable instructions. And, although the logical order is shown in the flowcharts, in some cases the steps shown or described may be performed in a different order than the ones described herein.
  • FIG. 1 is a pooling method of image information according to an embodiment of the present application. As shown in FIG. 1 , the method includes the following steps:
  • Step S102 Acquire at least one image information to be processed in the pooling window.
  • At least one image information to be processed may be an output result of a convolution operation.
  • the pooling method of the image information in this embodiment further includes the following steps:
  • Step S10 detecting whether at least one image information to be processed completes a convolution operation.
  • step S10 of the present application after acquiring at least one image to be processed in the pooling window, first, it may be determined whether the at least one image information to be processed has completed the convolution operation.
  • Step S12 Perform performing a first pooling process on at least one image information to be processed if at least one image information to be processed completes a convolution operation.
  • step S12 of the present application if at least one image information to be processed completes the convolution operation, step 104 is performed.
  • Step S14 If there is image information to be processed in which the convolution operation is not completed in at least one image information to be processed, the image to be processed in which at least one of the image information to be processed has been subjected to the convolution operation is subjected to the first pooling process to obtain a partial pool.
  • the result of the buffering is buffered to the buffer; when there is new image information to be processed in the pooling window to complete the convolution operation, the partial pooling result stored in the buffer and the new image information to be processed are performed. Two pools of processing.
  • Step S104 Perform a first pooling process on the at least one image to be processed to obtain at least one pooling result.
  • the first pooling process may include a maximum pooling operation or an average pooling operation.
  • the at least one image information to be processed may be subjected to a first pooling process to obtain at least one pooling result.
  • Step S106 buffering at least one pooling result to the buffer.
  • the buffer is a first in first out buffer.
  • the at least one pooling result is buffered into the buffer, and the pooling window moves a step in each operation cycle, and in a certain operation cycle, the pooled window is read.
  • the pooling window moves by one step, and the reading pooling window is not given the previous one.
  • the image information to be processed is repeated in the operation cycle, and the pooled result corresponding to the new image information to be processed is stored in the buffer to replace the pooling result of the previous round.
  • Step S108 performing a second pooling process on the at least one previous pooling result and the at least one pooling result in the buffer, to obtain a pooling result of the pooling window.
  • step S108 of the present application after buffering at least one pooling result to the buffer, reading at least one pooling result in the buffer and at least one previous pooling result, and at least one pooling result and at least one previous The pooling result is subjected to the second pooling process, and the final pooling result of the pooling window has been obtained.
  • the second pooling process may include a maximum pooling operation or an average pooling operation.
  • the pooling method of the image information in the embodiment further includes: : determining whether the number of pooled results in the buffer is greater than or equal to a preset value; and if the number of pooled results buffered in the buffer is greater than a preset value, performing a second pooling process.
  • the stride of the buffer is N
  • the at least one pooling result after the first pooling process is cached to the buffer, and then all the pooled results stored in the buffer are subjected to the second pooling operation to obtain the final pooling result. It achieves the purpose of pooling processing without waiting for the convolutional layer to complete all operations, thereby achieving the technical effect of reducing the read bandwidth, and thus solving the problem that all data waiting for the convolution layer is completed.
  • the method before acquiring at least one image information to be processed in the pooling window, the method further includes: performing convolution operation on at least one image information to be processed to obtain a convolution operation result; and storing a convolution operation result.
  • the method before performing convolution operation on the at least one image to be processed, the method further includes: acquiring the original image; converting the original image into at least one image to be processed; according to the size of the at least one image to be processed, Obtain at least one image information to be processed.
  • the acquiring at least one image information to be processed according to the size of the at least one image to be processed includes: if the size of the image to be processed is smaller than the preset value, extracting the contour information from the image to be processed as the image information to be processed; The size of the processed image is greater than or equal to the preset value, and the detailed information is extracted as the image information to be processed in the image to be processed.
  • a piece of the original image is gradually transformed into N different sizes of the image to be processed after being changed by the N level, and then the two-dimensional information is sampled and extracted for the different images to be processed (image position, image) Information), for the small-sized image to be processed, the contour information can be extracted, the contour of the image is removed in the large-sized image to be processed, and the detail information is retained, and the large-sized image to be processed is subjected to the same type of two-dimensional Sample extraction.
  • the original image is divided into different levels for convolution pooling, so that a faster transformation speed can be realized by convolution transformation of different parts, and the transformation speed of the image can be performed when the hardware can synchronously process different levels of transformation.
  • the double speed is improved, and the features of the image can be extracted more accurately by using different convolution templates for different contours.
  • the embodiment of the present application further provides a pooling device for image information. It should be noted that the pooling device of the image information of this embodiment may be used to perform the pooling method of the image information in the embodiment of the present application.
  • the pooling device of the image information includes an obtaining unit 20, a processing unit 22, and a buffer unit 24.
  • the obtaining unit 20 is configured to obtain at least one image to be processed in the pooling window
  • the processing unit 22 is configured to perform a first pooling process on the at least one image to be processed to obtain at least one pooling result;
  • a cache unit 24 configured to cache at least one pooled result to a buffer
  • the processing unit 22 is further configured to perform a second pooling process on the at least one previous pooling result and the at least one pooling result in the buffer to obtain a pooling result of the pooling window.
  • the processing unit 22 is further configured to determine whether the number of pooled results buffered in the buffer is greater than or equal to a preset value; if the number of pooled results buffered in the buffer is greater than a preset value , the second pooling process.
  • the processing unit 22 is further configured to: detect whether the at least one image information to be processed completes the convolution operation; if the at least one image information to be processed completes the convolution operation, perform a first pooling process on the at least one image information to be processed. If at least one image information to be processed of the uncompleted convolution operation exists in at least one image information to be processed, the image to be processed of the at least one convolution operation in the image information to be processed is subjected to the first pooling process to obtain a partial pooling result. Cache the partial pooling result to the buffer; when there is new image information to be processed in the pooling window to complete the convolution operation, the partial pooling result stored in the buffer and the new image information to be processed are performed in the second pool. Processing.
  • the processing unit 22 is further configured to perform a convolution operation on the at least one image information to be processed to obtain a convolution operation result; and store a convolution operation result.
  • the obtaining unit 20 is further configured to acquire an original image; the processing unit 22 is further configured to convert the original image into at least one image to be processed; and acquire at least one image information to be processed according to the size of the at least one image to be processed.
  • the processing unit 22 is configured to: obtain, according to the size of the at least one image to be processed, at least one image information to be processed: if the size of the image to be processed is smaller than a preset value, extract contour information from the image to be processed as The image information to be processed; if the size of the image to be processed is greater than or equal to the preset value, the detail information is extracted as the image information to be processed in the image to be processed.
  • At least one image to be processed in the pooling window is acquired; performing a first pooling process on the at least one image to be processed to obtain at least one pooling result; and buffering at least one pooling result into the buffer And performing a second pooling process on the at least one previous pooling result in the buffer and the at least one pooling result, and obtaining the pooling result of the pooling window by first processing the first pooled at least one pool
  • the result is cached to the buffer, and then all the pooled results stored in the buffer are subjected to the second pooling operation to obtain the final pooling result, so that the pool can be performed without waiting for the convolution layer to complete all the operations.
  • the purpose of the processing is to achieve the technical effect of reducing the read bandwidth, and further solve the problem that the input of the pooled layer to the pooling process is required to be occupied because all the data of the convolution layer needs to be completed.
  • the technical problem of reading bandwidth is to achieve the technical effect of reducing the read bandwidth, and further solve the problem that the input of the pooled layer to the pooling process is required to be occupied because all the data of the convolution layer needs to be completed.
  • the embodiment of the present application further provides a pooling device for image information.
  • the pooling device of the image information includes: a pooling window; a processor, a processor running program, wherein, when the program is running, the following processing steps are performed on the data output from the pooling window: acquiring at least one image to be processed in the pooling window Performing a first pooling process on at least one image information to be processed to obtain at least one pooling result; buffering at least one pooling result to the buffer; at least one previous pooling result in the buffer and at least one pooling result The second pooling process is performed to obtain the pooling result of the pooling window.
  • the embodiment of the present application further provides a storage medium including a stored program, wherein the program executes a pooling method of image information having any of the above features.
  • the embodiment of the present application further provides a processor, where the processor is used to run a program, wherein the program performs a pooling method for image information having any of the above features.
  • the disclosed technical contents may be implemented in other manners.
  • the device embodiments described above are only schematic.
  • the division of the unit may be a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, unit or module, and may be electrical or otherwise.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium.
  • a computer readable storage medium A number of instructions are included to cause a computer device (which may be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present application.
  • the foregoing storage medium includes: a U disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk, and the like. .

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Image Processing (AREA)

Abstract

本申请公开了一种图像信息的池化方法及装置、存储介质、处理器。其中，该方法包括：获取池化窗内至少一个待处理图像信息；对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将至少一个池化结果缓存至缓冲器；对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果。本申请解决了由于需等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理造成的需要占用较大的读取带宽的技术问题。

Description

图像信息的池化方法及装置、存储介质、处理器 技术领域
本申请涉及图像处理领域，具体而言，涉及一种图像信息的池化方法及装置、存储介质、处理器。
背景技术
目前随着人工智能的火热发展，对于图像的处理需求正在逐步的增大。在进行图像信息处理的过程中，传统上对多个图像信息进行池化处理时，需要等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理，因此，需要占用较大的读取带宽。
针对上述的问题，目前尚未提出有效的解决方案。
发明内容
本申请实施例提供了一种图像信息的池化方法及装置、存储介质、处理器，以至少解决由于需等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理造成的需要占用较大的读取带宽的技术问题。
根据本申请实施例的一个方面，提供了一种图像信息的池化方法，包括：获取池化窗内至少一个待处理图像信息；对上述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将上述至少一个池化结果缓存至缓冲器；对上述缓冲器中的至少一个先前池化结果以及上述至少一个池化结果进行第二池化处理，得到上述池化窗的池化结果。
可选地，对上述缓冲器中的至少一个先前池化结果以及上述至少一个池化结果进行第二池化处理，得到上述池化窗的池化结果之前，上述方法还包括：判断上述缓冲器中缓存的池化结果的个数是否大于等于预设值；在上述缓冲器中缓存的池化结果的个数大于等上述预设值的情况下，进行上述第二池化处理。
可选地，获取池化窗内至少一个待处理图像信息之后，上述方法还包括：检测上述至少一个待处理图像信息是否完成卷积运算；若上述至少一个待处理图像信息完成卷积运算，执行上述对上述至少一个待处理图像信息进行第一池化处理；若上述至少 一个待处理图像信息中存在未完成卷积运算的待处理图像信息，将上述至少一个待处理图像信息中已完成卷积运算的待处理图像进行上述第一池化处理，得到部分池化结果；将上述部分池化结果缓存至上述缓冲器；当上述池化窗中有新的待处理图像信息完成卷积运算时，将上述缓冲器所储存的上述部分池化结果及新的待处理图像信息进行上述第二池化处理。
可选地，获取池化窗内至少一个待处理图像信息之前，上述方法还包括：对上述至少一个待处理图像信息进行卷积运算，得到卷积运算结果；存储上述卷积运算结果。
可选地，对上述至少一个待处理图像信息进行卷积运算之前，上述方法还包括：获取原始图像；将上述原始图像转化为至少一个待处理图像；根据上述至少一个待处理图像的尺寸，获取上述至少一个待处理图像信息。
可选地，上述根据上述至少一个待处理图像的尺寸，获取上述至少一个待处理图像信息包括：若上述待处理图像的尺寸小于预设值，则从上述待处理图像中提取轮廓信息作为上述待处理图像信息；若上述待处理图像的尺寸大于等于上述预设值，则上述待处理图像中提取细节信息作为上述待处理图像信息。
根据本申请实施例的另一方面，还提供了一种图像信息的池化装置，包括：获取单元，用于获取池化窗内至少一个待处理图像信息；处理单元，用于对上述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；缓存单元，用于将上述至少一个池化结果缓存至缓冲器；上述处理单元，还用于对上述缓冲器中的至少一个先前池化结果以及上述至少一个池化结果进行第二池化处理，得到上述池化窗的池化结果。
根据本申请实施例的另一方面，还提供了一种图像信息的池化装置，包括：池化窗；处理器，上述处理器运行程序，其中，上述程序运行时对于从上述池化窗输出的数据执行如下处理步骤：获取上述池化窗内至少一个待处理图像信息；对上述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将上述至少一个池化结果缓存至缓冲器；对上述缓冲器中的至少一个先前池化结果以及上述至少一个池化结果进行第二池化处理，得到上述池化窗的池化结果。
根据本申请实施例的另一方面，还提供了一种存储介质，上述存储介质包括存储的程序，其中，上述程序执行具有上述任意特征的图像信息的池化方法。
根据本申请实施例的另一方面，还提供了一种处理器，上述处理器用于运行程序，其中，上述程序运行时执行具有上述任意特征的图像信息的池化方法。
在本申请实施例中，采用获取池化窗内至少一个待处理图像信息；对至少一个待 处理图像信息进行第一池化处理，得到至少一个池化结果；将至少一个池化结果缓存至缓冲器；对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果的方式，通过先将第一池化处理后的至少一个池化结果缓存至缓冲器，再将所有存储在缓冲器的池化结果进行第二次池化运算，以得到最终的池化结果，达到了无需等待卷积层将所有数据完成运算即可进行池化处理的目的，从而实现了降低读取带宽的技术效果，进而解决了由于需等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理造成的需要占用较大的读取带宽的技术问题。
附图说明
此处所说明的附图用来提供对本申请的进一步理解，构成本申请的一部分，本申请的示意性实施例及其说明用于解释本申请，并不构成对本申请的不当限定。在附图中：
图1是根据本申请实施例的一种可选的图像信息的池化方法的流程示意图；
图2是根据本申请实施例的一种可选的图像信息的池化装置的结构示意图。
具体实施方式
为了使本技术领域的人员更好地理解本申请方案，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分的实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本申请保护的范围。
需要说明的是，本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的本申请的实施例能够以除了在这里图示或描述的那些以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
实施例1
根据本申请实施例，提供了一种图像信息的池化方法的方法实施例，需要说明的是，在附图的流程图示出的步骤可以在诸如一组计算机可执行指令的计算机系统中执行，并且，虽然在流程图中示出了逻辑顺序，但是在某些情况下，可以以不同于此处的顺序执行所示出或描述的步骤。
图1是根据本申请实施例的图像信息的池化方法，如图1所示，该方法包括如下步骤：
步骤S102，获取池化窗内至少一个待处理图像信息。
本申请上述步骤S102中，至少一个待处理图像信息可以为卷积运算的输出结果。可选地，获取池化窗内至少一个待处理图像信息之后，本实施例的图像信息的池化方法还包括以下步骤：
步骤S10，检测至少一个待处理图像信息是否完成卷积运算。
本申请上述步骤S10中，在获取到池化窗内至少一个待处理图像信息之后，首先，可以判断该至少一个待处理图像信息是否已完成卷积运算。
步骤S12，若至少一个待处理图像信息完成卷积运算，执行对至少一个待处理图像信息进行第一池化处理。
本申请上述步骤S12中，若至少一个待处理图像信息完成卷积运算，则执行步骤104。
步骤S14，若至少一个待处理图像信息中存在未完成卷积运算的待处理图像信息，将至少一个待处理图像信息中已完成卷积运算的待处理图像进行第一池化处理，得到部分池化结果；将部分池化结果缓存至缓冲器；当池化窗中有新的待处理图像信息完成卷积运算时，将缓冲器所储存的部分池化结果及新的待处理图像信息进行第二池化处理。
步骤S104，对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果。
本申请上述步骤S104中，第一池化处理可以包括最大池化运算或平均池化运算。在获取到至少一个待处理图像信息之后，可以对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果。
步骤S106，将至少一个池化结果缓存至缓冲器。
本申请上述步骤S106中，缓冲器为先进先出缓冲器。在得到至少一个池化结果之 后，将该至少一个池化结果缓存至缓冲器内，池化窗在每个运算周期都会移动一个步幅，在某一个运算周期中，读取池化窗内的至少一个待处理图像信息，以进行第一池化运算，并将对应的池化结果存储在缓冲器中，在下一个运算周期，池化窗移动一个步幅，读取池化窗不予前一个运算周期重复的待处理图像信息，并将新的待处理图像信息对应的池化结果存储在缓冲器中，以取代前一轮的池化结果。
步骤S108，对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果。
本申请上述步骤S108中，在将至少一个池化结果缓存至缓冲器之后，读取缓冲器内的至少一个池化结果以及至少一个先前池化结果，并对至少一个池化结果以及至少一个先前池化结果进行第二池化处理，已得到池化窗的最终池化结果。其中，第二池化处理可以包括最大池化运算或平均池化运算。
可选地，对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果之前，本实施例的图像信息的池化方法还包括：判断缓冲器中缓存的池化结果的个数是否大于等于预设值；在缓冲器中缓存的池化结果的个数大于等预设值的情况下，进行第二池化处理。
举例而言，但缓冲器的步幅为N时，则需要等待缓冲器存储了N个池化结果后，进行第二池化处理。
通过上述步骤，先将第一池化处理后的至少一个池化结果缓存至缓冲器，再将所有存储在缓冲器的池化结果进行第二次池化运算，以得到最终的池化结果，达到了无需等待卷积层将所有数据完成运算即可进行池化处理的目的，从而实现了降低读取带宽的技术效果，进而解决了由于需等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理造成的需要占用较大的读取带宽的技术问题。
作为一种可选的实现方式，获取池化窗内至少一个待处理图像信息之前，方法还包括：对至少一个待处理图像信息进行卷积运算，得到卷积运算结果；存储卷积运算结果。
作为一种可选的实现方式，对至少一个待处理图像信息进行卷积运算之前，方法还包括：获取原始图像；将原始图像转化为至少一个待处理图像；根据至少一个待处理图像的尺寸，获取至少一个待处理图像信息。
可选的，根据至少一个待处理图像的尺寸，获取至少一个待处理图像信息包括：若待处理图像的尺寸小于预设值，则从待处理图像中提取轮廓信息作为待处理图像信息；若待处理图像的尺寸大于等于预设值，则待处理图像中提取细节信息作为待处理 图像信息。
本实施例中，通过将一张原始图像，经过N等级的变化，逐渐的变换成N个不同尺寸的待处理图像，然后对不同的待处理图像进行二维信息的采样提取(图像位置，图像信息)，对于小尺寸的待处理图像可以进行轮廓信息的提取，对于大尺寸的待处理图像中将图像的轮廓进行剔除，而对细节信息进行保留，对大尺寸的待处理图像进行同类二维采样提取。
进而，将原始图像分成不同的层次进行卷积池化，这样通过对不同部分进行卷积变换可以实现更快的变换速度，当硬件可以同步的处理不同等级的变换时，可以对图像的变换速度实现倍速的提升，同时由于对不同的轮廓使用不同的卷积模板可以更加精准的提取图像的特征。
实施例2
本申请实施例还提供了一种图像信息的池化装置。需要说明的是，该实施例的图像信息的池化装置可以用于执行本申请实施例的图像信息的池化方法。
图2是根据本申请实施例的一种图像信息处理装置的示意图。如图2所示，该图像信息的池化装置包括：获取单元20、处理单元22和缓存单元24。
获取单元20，用于获取池化窗内至少一个待处理图像信息；
处理单元22，用于对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；
缓存单元24，用于将至少一个池化结果缓存至缓冲器；
处理单元22，还用于对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果。
可选地，处理单元22，还用于判断缓冲器中缓存的池化结果的个数是否大于等于预设值；在缓冲器中缓存的池化结果的个数大于等预设值的情况下，进行第二池化处理。
可选地，处理单元22，还用于检测至少一个待处理图像信息是否完成卷积运算；若至少一个待处理图像信息完成卷积运算，执行对至少一个待处理图像信息进行第一池化处理；若至少一个待处理图像信息中存在未完成卷积运算的待处理图像信息，将 至少一个待处理图像信息中已完成卷积运算的待处理图像进行第一池化处理，得到部分池化结果；将部分池化结果缓存至缓冲器；当池化窗中有新的待处理图像信息完成卷积运算时，将缓冲器所储存的部分池化结果及新的待处理图像信息进行第二池化处理。
可选地，处理单元22，还用于对至少一个待处理图像信息进行卷积运算，得到卷积运算结果；存储卷积运算结果。
可选地，获取单元20，还用于获取原始图像；处理单元22，还用于将原始图像转化为至少一个待处理图像；根据至少一个待处理图像的尺寸，获取至少一个待处理图像信息。
可选地，处理单元22用于执行以下步骤根据至少一个待处理图像的尺寸，获取至少一个待处理图像信息：若待处理图像的尺寸小于预设值，则从待处理图像中提取轮廓信息作为待处理图像信息；若待处理图像的尺寸大于等于预设值，则待处理图像中提取细节信息作为待处理图像信息。
在本申请实施例中，采用获取池化窗内至少一个待处理图像信息；对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将至少一个池化结果缓存至缓冲器；对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果的方式，通过先将第一池化处理后的至少一个池化结果缓存至缓冲器，再将所有存储在缓冲器的池化结果进行第二次池化运算，以得到最终的池化结果，达到了无需等待卷积层将所有数据完成运算即可进行池化处理的目的，从而实现了降低读取带宽的技术效果，进而解决了由于需等待卷积层的所有数据均完成运算，才会输入至池化层进行池化处理造成的需要占用较大的读取带宽的技术问题。
实施例3
本申请实施例还提供了一种图像信息的池化装置。该图像信息的池化装置包括：池化窗；处理器，处理器运行程序，其中，程序运行时对于从池化窗输出的数据执行如下处理步骤：获取池化窗内至少一个待处理图像信息；对至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将至少一个池化结果缓存至缓冲器；对缓冲器中的至少一个先前池化结果以及至少一个池化结果进行第二池化处理，得到池化窗的池化结果。
本申请实施例还提供了一种存储介质，存储介质包括存储的程序，其中，程序执 行具有上述任意特征的图像信息的池化方法。
本申请实施例还提供了一种处理器，处理器用于运行程序，其中，程序执行具有上述任意特征的图像信息的池化方法。
上述本申请实施例序号仅仅为了描述，不代表实施例的优劣。
在本申请的上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的技术内容，可通过其它的方式实现。其中，以上所描述的装置实施例仅仅是示意性的，例如所述单元的划分，可以为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，单元或模块的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述仅是本申请的优选实施方式，应当指出，对于本技术领域的普通技术人员来说，在不脱离本申请原理的前提下，还可以做出若干改进和润饰，这些改进和润饰也应视为本申请的保护范围。

Claims (10)

  1. 一种图像信息的池化方法，包括：
    获取池化窗内至少一个待处理图像信息；
    对所述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；
    将所述至少一个池化结果缓存至缓冲器；
    对所述缓冲器中的至少一个先前池化结果以及所述至少一个池化结果进行第二池化处理，得到所述池化窗的池化结果。
  2. 根据权利要求1所述的方法，其中，对所述缓冲器中的至少一个先前池化结果以及所述至少一个池化结果进行第二池化处理，得到所述池化窗的池化结果之前，所述方法还包括：
    判断所述缓冲器中缓存的池化结果的个数是否大于等于预设值；
    在所述缓冲器中缓存的池化结果的个数大于等所述预设值的情况下，进行所述第二池化处理。
  3. 根据权利要求1所述的方法，其中，获取池化窗内至少一个待处理图像信息之后，所述方法还包括：
    检测所述至少一个待处理图像信息是否完成卷积运算；
    若所述至少一个待处理图像信息完成卷积运算，执行所述对所述至少一个待处理图像信息进行第一池化处理；
    若所述至少一个待处理图像信息中存在未完成卷积运算的待处理图像信息，将所述至少一个待处理图像信息中已完成卷积运算的待处理图像进行所述第一池化处理，得到部分池化结果；
    将所述部分池化结果缓存至所述缓冲器；
    当所述池化窗中有新的待处理图像信息完成卷积运算时，将所述缓冲器所储存的所述部分池化结果及新的待处理图像信息进行所述第二池化处理。
  4. 根据权利要求1所述的方法，其中，获取池化窗内至少一个待处理图像信息之前，所述方法还包括：
    对所述至少一个待处理图像信息进行卷积运算，得到卷积运算结果；
    存储所述卷积运算结果。
  5. 根据权利要求4所述的方法，其中，对所述至少一个待处理图像信息进行卷积运算之前，所述方法还包括：
    获取原始图像；
    将所述原始图像转化为至少一个待处理图像；
    根据所述至少一个待处理图像的尺寸，获取所述至少一个待处理图像信息。
  6. 根据权利要求5所述的方法，其中，所述根据所述至少一个待处理图像的尺寸，获取所述至少一个待处理图像信息包括：
    若所述待处理图像的尺寸小于预设值，则从所述待处理图像中提取轮廓信息作为所述待处理图像信息；
    若所述待处理图像的尺寸大于等于所述预设值，则所述待处理图像中提取细节信息作为所述待处理图像信息。
  7. 一种图像信息的池化装置，包括：
    获取单元，用于获取池化窗内至少一个待处理图像信息；
    处理单元，用于对所述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；
    缓存单元，用于将所述至少一个池化结果缓存至缓冲器；
    所述处理单元，还用于对所述缓冲器中的至少一个先前池化结果以及所述至少一个池化结果进行第二池化处理，得到所述池化窗的池化结果。
  8. 一种图像信息的池化装置，包括：
    池化窗；
    处理器，所述处理器运行程序，其中，所述程序运行时对于从所述池化窗输出的数据执行如下处理步骤：获取所述池化窗内至少一个待处理图像信息；对所述至少一个待处理图像信息进行第一池化处理，得到至少一个池化结果；将所述至少一个池化结果缓存至缓冲器；对所述缓冲器中的至少一个先前池化结果以及所述至少一个池化结果进行第二池化处理，得到所述池化窗的池化结果。
  9. 一种存储介质，所述存储介质包括存储的程序，其中，所述程序执行权利要求1至6中任意一项所述的图像信息的池化方法。
  10. 一种处理器，所述处理器用于运行程序，其中，所述程序运行时执行权利要求1至6中任意一项所述的图像信息的池化方法。
PCT/CN2018/099597 2017-10-19 2018-08-09 图像信息的池化方法及装置、存储介质、处理器 WO2019076109A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710983504.3A CN107749044A (zh) 2017-10-19 2017-10-19 图像信息的池化方法及装置
CN201710983504.3 2017-10-19

Publications (1)

Publication Number Publication Date
WO2019076109A1 true WO2019076109A1 (zh) 2019-04-25

Family

ID=61252942

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/099597 WO2019076109A1 (zh) 2017-10-19 2018-08-09 图像信息的池化方法及装置、存储介质、处理器

Country Status (2)

Country Link
CN (1) CN107749044A (zh)
WO (1) WO2019076109A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112308762A (zh) * 2020-10-23 2021-02-02 北京三快在线科技有限公司 一种数据处理方法及装置

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107749044A (zh) * 2017-10-19 2018-03-02 珠海格力电器股份有限公司 图像信息的池化方法及装置
CN110322388B (zh) * 2018-03-29 2023-09-12 上海熠知电子科技有限公司 池化方法及装置、池化系统、计算机可读存储介质
CN110322389B (zh) * 2018-03-29 2023-03-21 上海熠知电子科技有限公司 池化方法、装置及系统、计算机可读存储介质
CN110399977A (zh) * 2018-04-25 2019-11-01 华为技术有限公司 池化运算装置
CN110383330A (zh) * 2018-05-30 2019-10-25 深圳市大疆创新科技有限公司 池化装置和池化方法
CN111429334A (zh) * 2020-03-26 2020-07-17 光子算数(北京)科技有限责任公司 一种数据处理方法、装置、存储介质及电子设备

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140288928A1 (en) * 2013-03-25 2014-09-25 Gerald Bradley PENN System and method for applying a convolutional neural network to speech recognition
CN106779060A (zh) * 2017-02-09 2017-05-31 武汉魅瞳科技有限公司 一种适于硬件设计实现的深度卷积神经网络的计算方法
CN106981067A (zh) * 2017-04-05 2017-07-25 深圳市唯特视科技有限公司 一种基于全卷积网络的纹理分割方法
CN107749044A (zh) * 2017-10-19 2018-03-02 珠海格力电器股份有限公司 图像信息的池化方法及装置

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9978014B2 (en) * 2013-12-18 2018-05-22 Intel Corporation Reconfigurable processing unit
CN106228240B (zh) * 2016-07-30 2020-09-01 复旦大学 基于fpga的深度卷积神经网络实现方法
CN107239824A (zh) * 2016-12-05 2017-10-10 北京深鉴智能科技有限公司 用于实现稀疏卷积神经网络加速器的装置和方法
CN106875012B (zh) * 2017-02-09 2019-09-20 武汉魅瞳科技有限公司 一种基于fpga的深度卷积神经网络的流水化加速系统
CN106940815B (zh) * 2017-02-13 2020-07-28 西安交通大学 一种可编程卷积神经网络协处理器ip核

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20140288928A1 (en) * 2013-03-25 2014-09-25 Gerald Bradley PENN System and method for applying a convolutional neural network to speech recognition
CN106779060A (zh) * 2017-02-09 2017-05-31 武汉魅瞳科技有限公司 一种适于硬件设计实现的深度卷积神经网络的计算方法
CN106981067A (zh) * 2017-04-05 2017-07-25 深圳市唯特视科技有限公司 一种基于全卷积网络的纹理分割方法
CN107749044A (zh) * 2017-10-19 2018-03-02 珠海格力电器股份有限公司 图像信息的池化方法及装置

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112308762A (zh) * 2020-10-23 2021-02-02 北京三快在线科技有限公司 一种数据处理方法及装置

Also Published As

Publication number Publication date
CN107749044A (zh) 2018-03-02

Similar Documents

Publication Publication Date Title
WO2019076109A1 (zh) 图像信息的池化方法及装置、存储介质、处理器
TWI773189B (zh) 基於人工智慧的物體檢測方法、裝置、設備及儲存媒體
WO2021017261A1 (zh) 识别模型训练方法、图像识别方法、装置、设备及介质
CN108664981B (zh) 显著图像提取方法及装置
WO2016054779A1 (en) Spatial pyramid pooling networks for image processing
EP3678088B1 (en) Method and system of convolution in neural network with variable dilation rate
WO2022247005A1 (zh) 图像中目标物识别方法、装置、电子设备及存储介质
CN111860398B (zh) 遥感图像目标检测方法、系统及终端设备
WO2015184992A1 (zh) 一种识别重复图片的方法、图片搜索去重方法及其装置
JP2020135872A (ja) 物体認識方法、装置及びネットワーク
WO2017116916A1 (en) Image integration search based on human visual pathway model
WO2019119396A1 (zh) 人脸表情识别方法及装置
WO2021082819A1 (zh) 一种图像生成方法、装置及电子设备
WO2022242122A1 (zh) 一种视频优化方法、装置、终端设备及存储介质
WO2019127102A1 (zh) 信息处理方法、装置、云处理设备以及计算机程序产品
CN110399826B (zh) 一种端到端人脸检测和识别方法
TWI803243B (zh) 圖像擴增方法、電腦設備及儲存介質
US20240153240A1 (en) Image processing method, apparatus, computing device, and medium
CN110570394A (zh) 医学图像分割方法、装置、设备及存储介质
CN114758145B (zh) 一种图像脱敏方法、装置、电子设备及存储介质
WO2017202086A1 (zh) 图片的筛选方法及装置
US9171227B2 (en) Apparatus and method extracting feature information of a source image
WO2018053710A1 (zh) 数字图像的形态学处理方法和数字图像处理装置
KR20230162010A (ko) 이미지들 및 비디오로부터 반사 특징들을 제거하기 위한 실시간 기계 학습-기반 프라이버시 필터
Vasconcelos et al. Cuf: Continuous upsampling filters

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18869331

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 18869331

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 30.09.2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18869331

Country of ref document: EP

Kind code of ref document: A1