WO2023050673A1

WO2023050673A1 - Image caching method and apparatus, and electronic device, storage medium and computer program product

Info

Publication number: WO2023050673A1
Application number: PCT/CN2022/074698
Authority: WO
Inventors: 王志宏
Original assignee: 上海商汤智能科技有限公司
Priority date: 2021-09-28
Filing date: 2022-01-28
Publication date: 2023-04-06
Also published as: CN113870093A

Abstract

Provided in the present disclosure are an image caching method and apparatus, and an electronic device, a storage medium and a computer program product. The method comprises: using each of a plurality of training processes to read a group of images, so as to obtain a plurality of groups of images, wherein the plurality of training processes correspond to the plurality of groups of images on a one-to-one basis; using a first training process in the plurality of training processes to apply for a shared memory corresponding to the plurality of groups of images, and sharing the shared memory, obtained by means of the application, to each of the plurality of training processes that is different from the first training process; and using each of the plurality of training processes to respectively cache, to the shared memory, the group of images read from the plurality of groups of images, such that each of the plurality of training processes reads the plurality of groups of images from the shared memory during the process of executing a neural network training step.

Description

Image caching method, device, electronic equipment, storage medium and computer program product

Cross References to Related Applications

This application is based on the Chinese patent application with the application number 202111145887.X, the application date is September 28, 2021, and the application name is "An Image Caching Method, Device, Electronic Equipment, and Readable Storage Medium", and requires the Chinese The priority of the patent application, the entire content of the Chinese patent application is hereby incorporated in this application by reference.

technical field

The present application relates to the technical field of computer vision, and in particular to an image caching method, device, electronic equipment, storage medium and computer program product.

Background technique

The training process of the neural network is divided into two parts: data processing and training. Among them, the data processing stage mainly includes the two steps of reading pictures from the hard disk and preprocessing the pictures. Usually, in order to improve the training speed, multiple processes are started on a physical machine to use multiple graphics cards for training at the same time.

At present, when the data of a single image is relatively large, it takes a very long time to read the image during the data processing stage, so it may happen that the training process uses the previous image to train the neural network, while the next image is still being read , data processing and training cannot be well run in parallel, and image reading takes too long, resulting in low training efficiency of the neural network.

Contents of the invention

Embodiments of the present disclosure expect to provide an image caching method, device, electronic equipment, storage medium, and computer program product.

The technical scheme of the embodiment of the present disclosure is realized in this way:

An embodiment of the present disclosure provides an image caching method, including:

Using each training process in multiple training processes to read a group of images to obtain multiple groups of images; wherein, the multiple training processes correspond to the multiple groups of images one-to-one;

Using the first training process among the plurality of training processes, apply for the shared memory corresponding to the plurality of groups of images, and share the applied shared memory to the plurality of training processes with the first training process different for each training process;

Using each training process in the plurality of training processes, respectively cache a group of images read in the plurality of groups of images into the shared memory for each training process in the plurality of training processes During the execution of the neural network training step, the plurality of sets of images are read from the shared memory.

In the above method, the use of each training process in multiple training processes to read a group of images to obtain multiple groups of images, including:

Using the second training process in the plurality of training processes, read the image path list recording the storage path of each image in the image data set, and broadcast the image path list to the plurality of training processes and the first training process Two different training processes for each training process;

Using each training process in the multiple training processes, based on the image path list, according to the corresponding preset image reading strategy, read a group of images from the image data set to obtain the multiple groups of images .

In the above method, before applying for the shared memory corresponding to the plurality of groups of images using the first training process in the plurality of training processes, the method further includes:

Using each training process in the plurality of training processes, calculate the memory size required to support the caching of a group of images read in the plurality of groups of images, and obtain a plurality of memories corresponding to the plurality of groups of images one-to-one. size;

Using the first training process, summarizing the multiple memory sizes to obtain an overall memory size that supports storage of the multiple groups of images;

Using the first training process in the multiple training processes to apply for the shared memory corresponding to the multiple groups of images includes:

Applying for the shared memory according to the overall memory size by using the first training process.

In the above method, the use of each training process in the plurality of training processes is used to calculate the memory size required to support the caching of the corresponding group of images in the plurality of groups of images, and to obtain the memory size corresponding to the plurality of groups of images. After the corresponding multiple memory sizes, the method also includes:

Using each training process in the multiple training processes to sum up the multiple memory sizes to obtain the overall memory size.

In the above method, the use of each training process in the plurality of training processes is used to calculate the memory size required to support the caching of a group of images read in the plurality of groups of images, and to obtain the memory size corresponding to the plurality of groups of images One-to-one correspondence with multiple memory sizes, including:

Using each training process in the plurality of training processes to obtain shape information of a group of images read in the plurality of groups of images;

Using each training process in the plurality of training processes, according to the shape information of a group of images read in the plurality of groups of images, calculate the memory size required by a group of images that support cache reading, and obtain the described Multiple memory sizes.

In the above method, after acquiring the shape information of a group of images read in the plurality of groups of images by using each of the plurality of training processes, the method further includes:

Using each training process in the plurality of training processes to summarize the shape information of the multiple groups of images to obtain an information summary result.

An embodiment of the present disclosure provides an image caching device, including:

The reading module is configured to use each training process in multiple training processes to read a group of images to obtain multiple groups of images; wherein, the multiple training processes correspond to the multiple groups of images one-to-one;

The processing module is configured to use the first training process in the plurality of training processes to apply for the shared memory corresponding to the plurality of groups of images, and share the applied for shared memory to the plurality of training processes and all each training session different from the first training session;

The caching module is configured to use each training process in the plurality of training processes to respectively cache a group of images read in the plurality of groups of images into the shared memory for the plurality of training processes Each training process reads the multiple sets of images from the shared memory during the execution of the neural network training step.

In the above device, the reading module is specifically configured to use the second training process in the plurality of training processes to read the image path list that records the storage path of each image in the image data set, and store the image path The list is broadcast to each training process different from the second training process in the plurality of training processes; using each training process in the plurality of training processes, based on the image path list, according to the corresponding preset image The reading strategy is to read a group of images from the image data set to obtain the multiple groups of images.

In the above device, the processing module is further configured to use each training process in the plurality of training processes to calculate the memory size required to support the caching of a group of images read in the plurality of groups of images, and obtain the same as Multiple memory sizes corresponding to the multiple groups of images one-to-one; using the first training process, summarizing the multiple memory sizes to obtain an overall memory size that supports storage of the multiple groups of images;

The processing module is specifically configured to use the first training process to apply for the shared memory according to the overall memory size.

In the above device, the processing module is further configured to use each training process in the multiple training processes to summarize the multiple memory sizes to obtain the overall memory size.

In the above device, the processing module is specifically configured to use each training process in the plurality of training processes to acquire the shape information of a group of images read in the plurality of groups of images; Each training process in the training process, according to the shape information of a group of images read in the plurality of groups of images, calculates the memory size required by a group of images that support cache reading, and obtains the multiple memory sizes.

In the above device, the processing module is further configured to use each training process in the multiple training processes to summarize the shape information of the multiple groups of images to obtain an information summary result.

An embodiment of the present disclosure provides an electronic device, and the electronic device includes: a processor, a memory, and a communication bus; wherein,

The communication bus is configured to realize connection and communication between the processor and the memory;

The processor is configured to execute one or more programs stored in the memory, so as to implement the above image caching method.

An embodiment of the present disclosure provides a computer-readable storage medium, the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to realize the above image caching method.

An embodiment of the present disclosure provides a computer program product, the computer program product includes a computer program or an instruction, and when the computer program or instruction is run on a computer, the computer is caused to execute the above image caching method.

Embodiments of the present disclosure provide an image caching method, device, electronic equipment, storage medium, and computer program product. The method includes: using each training process in multiple training processes to read a group of images to obtain multiple groups of images; Wherein, a plurality of training processes are in one-to-one correspondence with multiple groups of images; using the first training process in the multiple training processes, apply for shared memory corresponding to multiple groups of images, and share the applied shared memory to multiple training processes and Each training process that is different from the first training process; using each training process in the multiple training processes, respectively cache a group of images read in the multiple groups of images to the shared memory for each of the multiple training processes Each training process reads sets of images from shared memory during the neural network training steps. The technical solution provided by the embodiments of the present disclosure utilizes the training process to pre-cache the images to be applied during the neural network training phase into the shared memory, thereby increasing the image reading speed and improving the training efficiency of the neural network.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the background art, the following will describe the drawings that need to be used in the embodiments of the present disclosure or the background art.

The accompanying drawings here are incorporated into the description and constitute a part of the present description. These drawings show embodiments consistent with the present disclosure, and are used together with the description to explain the technical solution of the present disclosure.

FIG. 1 is a schematic flowchart of an image caching method provided by an embodiment of the present disclosure;

FIG. 2 is a schematic diagram of an exemplary training process cache image provided by an embodiment of the present disclosure;

FIG. 3 is a schematic structural diagram of an image caching device provided by an embodiment of the present disclosure;

Fig. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure.

Detailed ways

In order to make the purpose, technical solutions and advantages of the embodiments of the present disclosure clearer, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the described embodiments It is a part of the embodiments of the present disclosure, but not all of them. The following examples are used to illustrate the present disclosure, but not to limit the scope of the present disclosure. Based on the embodiments in the present disclosure, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present disclosure.

In the following description, references to "some embodiments" describe a subset of all possible embodiments, but it is understood that "some embodiments" may be the same subset or a different subset of all possible embodiments, and Can be combined with each other without conflict.

It should be pointed out that the term "first\second\third" involved in the embodiments of the present disclosure is only to distinguish similar objects, and does not represent a specific ordering of objects. Understandably, "first\second\third "Where permitted, the preset order or sequence may be interchanged so that the embodiments of the disclosure described herein can be practiced in an order other than that illustrated or described herein.

An embodiment of the present disclosure provides an image caching method, which may be executed by an image caching device. For example, the image caching method may be executed by a terminal device or a server or other electronic devices, wherein the terminal device may be a user equipment (User Equipment, UE), mobile devices, user terminals, cellular phones, cordless phones, personal digital assistants (Personal Digital Assistant, PDA), handheld devices, computing devices, vehicle-mounted devices, wearable devices, etc. In some possible implementation manners, the image caching method may be implemented by a processor invoking computer-readable instructions stored in a memory.

The embodiment of the present disclosure provides an image caching method. FIG. 1 is a schematic flowchart of an image caching method provided by an embodiment of the present disclosure. As shown in Figure 1, in the embodiment of the present disclosure, the image caching method mainly includes the following steps:

S101. Using each of the multiple training processes to read a set of images to obtain multiple sets of images; wherein, the multiple training processes are in one-to-one correspondence with the multiple sets of images.

In an embodiment of the present disclosure, the image caching device uses multiple training processes to read multiple sets of images.

It should be noted that, in the embodiment of the present disclosure, multiple sets of images are actually stored on the hard disk, and the image cache device can read a set of images by using each training process in multiple training processes, that is, multiple The training process is in one-to-one correspondence with multiple sets of images. The specific number of training processes and the images read by each training process can be set according to actual needs and application scenarios, which are not limited in this embodiment of the present disclosure.

It should be noted that, in the embodiments of the present disclosure, for each training process in the multiple training processes, a neural network training step may be performed for training the neural network.

It should be noted that, in the embodiments of the present disclosure, a group of images read by each training process may include one or more frames of images, and the specific number of images included in each group of images may be set according to actual needs. Certainly, the embodiments of the present disclosure are not limited.

Specifically, in an embodiment of the present disclosure, the image processing device uses each of the multiple training processes to read a set of images to obtain multiple sets of images, including: using the second training process among the multiple training processes, reading the image path list that records the storage path of each image in the image dataset, and broadcasting the image path list to each of the plurality of training processes that is different from the second training process; utilizing each of the plurality of training processes , based on the image path list, according to the corresponding preset image reading strategy, read a group of images from the image data set to obtain multiple groups of images.

It should be noted that, in the embodiments of the present disclosure, the multiple training processes include the first training process, wherein the first training process is used to realize the application and sharing of the shared memory, which will be detailed in the subsequent steps, and the second The second training process can be the same training process as the first training process, of course, it can also be any training process different from the first training process among multiple training processes. The specific second training process can be set according to actual needs and application scenarios, and is not limited in this embodiment of the present disclosure.

It can be understood that, in the embodiment of the present disclosure, the storage path of each image in the image data set is recorded in the image path list, and the image cache device uses the second training process to read the image path list and broadcast it to multiple training processes Each training process that is different from the second training process, thereby realizing that each training process in multiple training processes knows the storage path of each image in the image data set, so that the image cache device can use the storage path of each image in the multiple training processes Each training process reads a set of images based on the image path list.

It should be noted that, in the embodiments of the present disclosure, each of the multiple training processes is set with a corresponding preset image reading strategy, which indicates that the training thread needs to read a set of images from the image data set. group of images, that is, each training thread actually reads part of the images in the image dataset separately. For example, the multiple training processes are three training processes, specifically including: training process 1, training process 2, and training process 3, wherein the preset image reading strategy corresponding to training process 1 is to read the previous One third of the images, that is, the set of images read by training process 1 is the first third of the images recorded in the image path list in the image data set, and the preset image reading strategy corresponding to training process 2 is Read the middle third of the images recorded in the image path list, and the preset image reading strategy corresponding to training process 3 is to read the last third of the images recorded in the image path list, that is, use three training processes , which can realize the reading of all images in the image dataset. The specific preset image reading strategy corresponding to each training process can be set according to actual needs and application scenarios, and is not limited in this embodiment of the present disclosure.

It should be noted that, in the embodiments of the present disclosure, multiple sets of images may cover the entire image data set, and of course, may also cover only some specific images in the image data set, which is not limited in the embodiments of the present disclosure.

S102. Using the first training process among the multiple training processes, apply for shared memory corresponding to multiple groups of images, and share the applied shared memory to each training process different from the first training process among the multiple training processes .

In an embodiment of the present disclosure, when multiple sets of images are read by the image caching device using multiple training processes, use the first training process in the multiple training processes to apply for the shared memory corresponding to the multiple sets of images, and Sharing the applied for shared memory to each training process different from the first training process among the multiple training processes.

It should be noted that, in the embodiment of the present disclosure, the first training process may be any training process among multiple training processes, or may be a specific training process among multiple preset training processes, which may be related to The above-mentioned second training process is the same training process, and may also be a different training process from the above-mentioned second training process, which is not limited in this embodiment of the present disclosure.

It should be noted that, in the embodiment of the present disclosure, after the image caching device applies for the shared memory through the first training process, it can share the shared memory with other training processes. Specifically, the image caching device can use the first training process to send the descriptor pointing to the shared memory to other training processes, and the other training processes can learn the specific shared memory according to the descriptor.

Specifically, in the embodiment of the present disclosure, the image caching device may perform the following steps before applying for the shared memory corresponding to multiple groups of images using the first training process among the multiple training processes: using each of the multiple training processes to train process, calculate the memory size required to support caching a group of images read in multiple groups of images, and obtain multiple memory sizes corresponding to multiple groups of images; use the first training process to summarize multiple memory sizes, and obtain The overall memory size that supports storing multiple sets of images; correspondingly, the image cache device uses the first training process in multiple training processes to apply for shared memory corresponding to multiple sets of images, including: using the first training process, according to the overall memory size Apply for shared memory.

It can be understood that, in the embodiments of the present disclosure, when each training process reads a set of images, it can calculate the memory size required for caching the set of images, in fact, it is to calculate the set of images The image data size of the image.

It should be noted that, in the embodiment of the present disclosure, communication and interaction can be performed between multiple training processes, and each training process different from the first training process can use the calculated set of images that support cache reading Notify the memory size to the first training process, the first training process can summarize the multiple memory sizes corresponding to multiple sets of images, so as to obtain the overall memory size that supports storing multiple sets of images, and then apply for a share that matches the overall memory size Memory, that is, the shared memory corresponding to multiple sets of images. The overall memory size is actually the image data size of multiple sets of images.

Specifically, in an embodiment of the present disclosure, the image caching device uses each training process in multiple training processes to calculate the memory size required to support caching a group of images read in multiple groups of images, and obtain the Like one-to-one correspondence of multiple memory sizes, including: using each training process in multiple training processes to obtain the shape information of a set of images read in multiple sets of images; using each training process in multiple training processes , according to the shape information of a group of images read in multiple groups of images, calculate the memory size required by a group of images that support cache reading, and obtain multiple memory sizes.

It should be noted that, in the embodiment of the present disclosure, the image caching device can use each training process in multiple training processes to acquire the shape information of a group of images read, wherein, for a frame of image, its The shape information can include information such as the data type of the image, and each training process can calculate the memory size required by a group of images that support cache reading according to the shape information of a corresponding group of images.

Specifically, in an embodiment of the present disclosure, the image caching device uses each training process in multiple training processes to calculate the memory size required to support caching a group of images read in multiple groups of images, and obtain the After one-to-one correspondence of multiple memory sizes, the following steps can also be performed: use each training process in multiple training processes to aggregate multiple memory sizes to obtain the overall memory size.

It should be noted that, in the embodiment of the present disclosure, when the image caching device uses each training process to calculate the memory size required by a group of images that support cache reading, it can also use each training process to perform Aggregation of multiple memory sizes, so that each training process can learn the size of multiple sets of images.

Specifically, in an embodiment of the present disclosure, after the image caching device acquires the shape information of a set of images read from multiple sets of images by using each of the multiple training processes, the following steps may also be performed: respectively Using each training process in the multiple training processes, the shape information of multiple groups of images is summarized to obtain an information summary result.

It should be noted that, in the embodiment of the present disclosure, in the case that the image caching device obtains the shape information of a group of read images by using each training process, it can also use each training process to perform multiple sets of images respectively. Like the summary of shape information, so that each training process can learn the shape information of multiple sets of images, so that each training process can be based on the learned shape information when reading images from the shared memory. The selection of a specific image is then read.

S103. Using each training process in the multiple training processes, respectively cache a group of images read in multiple groups of images to the shared memory, so that each training process in the multiple training processes executes the neural network training step During the period, multiple sets of images are read from the shared memory.

In the embodiment of the present disclosure, after the image caching device uses the first training process to share the shared memory to other training processes in the multiple training processes, that is, the multiple training processes all know the shared memory, so that each of the multiple training processes can use Each training process caches a group of images read from multiple groups of images to the shared memory.

It can be understood that, in the embodiment of the present disclosure, the image caching device uses each training process in the multiple training processes to respectively cache a group of images read from the multiple groups of images into the shared memory, which is actually Realized that multiple groups of images are all cached in shared memory. For each training process, it knows the shared memory that caches multiple sets of images. Therefore, during the subsequent execution of the neural network training step, the image can be read from the shared memory. Specifically, the cache in the shared memory can be read Any image in the multiple groups of images is used for neural network training, thereby improving the image reading speed, and correspondingly, also improving the neural network training speed.

It should be noted that, in the embodiment of the present disclosure, the above-mentioned image caching method is preferably applicable to neural network training, and the amount of single-frame image data used for training is relatively large, but the total image data of the image is relatively small , that is, the application scenario with a small number of images.

FIG. 2 is a schematic diagram of an exemplary training process cache image provided by an embodiment of the present disclosure. As shown in Figure 2, training process 1 reads the image path list, and then broadcasts the path list to training process 2 and training process 3. Each training process determines a set of images that it needs to read according to the image path list, and reads a corresponding set of pictures from the hard disk. Each training process calculates the memory size required to cache these pictures in this process based on the read pictures, and then summarizes them, so that all training processes know the size of the entire image data set. Training process 1 applies for the corresponding shared memory according to the aggregated overall memory size, and shares this shared memory with other processes. Finally, each process caches the image it read before into the corresponding location of the shared memory, and each subsequent process can read the image in the shared memory during the training process.

An embodiment of the present disclosure provides an image caching method, including: using each training process in multiple training processes to read a set of images to obtain multiple sets of images; wherein, multiple training processes and multiple sets of images are one by one Corresponding; using the first training process in the multiple training processes, applying for shared memory corresponding to multiple groups of images, and sharing the applied shared memory to each training process different from the first training process in the multiple training processes; using multiple training processes Each training process in each training process caches a group of images read in multiple groups of images to the shared memory, so that each training process in multiple training processes can read from Read multiple sets of images in shared memory. The image caching method provided by the embodiments of the present disclosure utilizes the training process to pre-cache the images to be applied during the neural network training phase into the shared memory, thereby increasing the image reading speed and improving the neural network training efficiency.

An embodiment of the present disclosure provides an image caching device. FIG. 3 is a schematic structural diagram of an image caching device provided by an embodiment of the present disclosure. As shown in Figure 3, the image cache device includes:

The reading module 301 is configured to use each training process in multiple training processes to read a group of images to obtain multiple groups of images; wherein, the multiple training processes correspond to the multiple groups of images one by one;

The processing module 302 is configured to use the first training process in the multiple training processes to apply for the shared memory corresponding to the multiple groups of images, and share the applied shared memory to the multiple training processes and each training session different from the first training session;

The caching module 303 is configured to use each training process in the plurality of training processes to respectively cache a group of images read in the plurality of groups of images into the shared memory for the plurality of training Each training process in the process reads the multiple groups of images from the shared memory during the execution of the neural network training step.

In an embodiment of the present disclosure, the reading module 301 is specifically configured to use the second training process in the plurality of training processes to read the image path list that records the storage path of each image in the image data set, and The image path list is broadcast to each training process different from the second training process in the plurality of training processes; using each training process in the plurality of training processes, based on the image path list, according to the corresponding The preset image reading strategy reads a group of images from the image data set to obtain the multiple groups of images.

In an embodiment of the present disclosure, the processing module 302 is further configured to use each training process in the plurality of training processes to calculate the memory required to support caching a set of images read in the multiple sets of images size, to obtain a plurality of memory sizes corresponding to the plurality of groups of images; using the first training process, summarizing the plurality of memory sizes, and obtaining an overall memory size that supports storage of the plurality of groups of images;

The processing module 302 is specifically configured to use the first training process to apply for the shared memory according to the overall memory size.

In an embodiment of the present disclosure, the processing module 302 is further configured to use each training process in the multiple training processes to summarize the multiple memory sizes to obtain the overall memory size.

In an embodiment of the present disclosure, the processing module 302 is specifically configured to use each training process in the multiple training processes to acquire the shape information of a set of images read in the multiple sets of images; Each training process in the plurality of training processes, according to the shape information of a group of images read in the plurality of groups of images, calculates the memory size required by a group of images that support cache reading, and obtains the plurality of memory size.

In an embodiment of the present disclosure, the processing module 302 is further configured to respectively use each training process in the multiple training processes to summarize the shape information of the multiple groups of images to obtain an information summary result.

An embodiment of the present disclosure provides an electronic device. Fig. 4 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. As shown in FIG. 4, in an embodiment of the present disclosure, the electronic device includes: a processor 401, a memory 402, and a communication bus 403; wherein,

The communication bus 403 is configured to realize connection and communication between the processor 401 and the memory 402;

The processor 401 is configured to execute one or more programs stored in the memory 402 to implement the image caching method described above.

An embodiment of the present disclosure provides a computer-readable storage medium, the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to realize the above image caching method. The computer-readable storage medium can be a volatile memory (volatile memory), such as a random access memory (Random-Access Memory, RAM); or a non-volatile memory (non-volatile memory), such as a read-only memory (Read-Only Memory). -Only Memory, ROM), flash memory (flash memory), hard disk (Hard Disk Drive, HDD) or solid-state drive (Solid-State Drive, SSD); it can also be a respective device including one or any combination of the above memories, Such as mobile phones, computers, tablet devices, personal digital assistants, etc.

An embodiment of the present disclosure provides a computer program product, the computer program product includes a computer program or an instruction, and when the computer program or instruction is run on a computer, the computer is made to execute the above-mentioned image caching method.

Those skilled in the art should understand that the embodiments of the present disclosure may be provided as methods, systems, or computer program products. Accordingly, the present disclosure may take the form of a hardware embodiment, a software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present disclosure may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, optical storage, etc.) having computer-usable program code embodied therein.

The present disclosure is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present disclosure. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable signal processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable signal processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable signal processing device to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded into a computer or other programmable signal processing device, so that a series of operational steps are performed on the computer or other programmable device to produce computer-implemented processing, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above descriptions are only preferred embodiments of the present disclosure, and are not intended to limit the protection scope of the present disclosure.

Industrial Applicability

Embodiments of the present disclosure provide an image caching method, device, electronic equipment, storage medium, and computer program product, wherein the method includes: using each training process in multiple training processes to read a set of images to obtain multiple sets of images Image; among them, multiple training processes correspond to multiple groups of images one by one; use the first training process in multiple training processes to apply for shared memory corresponding to multiple groups of images, and share the applied shared memory to multiple training processes In each training process different from the first training process; each training process in multiple training processes is used to cache a group of images read in multiple groups of images to shared memory for multiple training processes Each training process in , reads multiple sets of images from shared memory during the neural network training steps. The technical solution provided by the embodiments of the present disclosure utilizes the training process to pre-cache the images to be applied during the neural network training phase into the shared memory, thereby increasing the image reading speed and improving the training efficiency of the neural network.

Claims

An image caching method, comprising:

Using each training process in multiple training processes to read a group of images to obtain multiple groups of images; wherein, the multiple training processes correspond to the multiple groups of images one-to-one;

Using the first training process among the plurality of training processes, apply for the shared memory corresponding to the plurality of groups of images, and share the applied shared memory to the plurality of training processes with the first training process different for each training process;

Using each training process in the plurality of training processes, respectively cache a group of images read in the plurality of groups of images into the shared memory for each training process in the plurality of training processes During the execution of the neural network training step, the plurality of sets of images are read from the shared memory.
The method according to claim 1, wherein said using each training process in multiple training processes to read a group of images to obtain multiple groups of images comprises:

Using the second training process in the plurality of training processes, read the image path list recording the storage path of each image in the image data set, and broadcast the image path list to the plurality of training processes and the first training process Two different training processes for each training process;

Using each training process in the multiple training processes, based on the image path list, according to the corresponding preset image reading strategy, read a group of images from the image data set to obtain the multiple groups of images .
The method according to claim 1, wherein, before applying for the shared memory corresponding to the plurality of groups of images using the first training process in the plurality of training processes, the method further comprises:

Using each training process in the plurality of training processes, calculate the memory size required to support the caching of a group of images read in the plurality of groups of images, and obtain a plurality of memories corresponding to the plurality of groups of images one-to-one. size;

Using the first training process, summarizing the multiple memory sizes to obtain an overall memory size that supports storage of the multiple groups of images;

Using the first training process in the multiple training processes to apply for the shared memory corresponding to the multiple groups of images includes:

Applying for the shared memory according to the overall memory size by using the first training process.
The method according to claim 3, wherein, using each training process in the plurality of training processes, calculating the memory size required to support caching a corresponding group of images in the plurality of groups of images, and obtaining the same size as the After multiple memory sizes corresponding to multiple groups of images one-to-one, the method also includes:

Using each training process in the multiple training processes to sum up the multiple memory sizes to obtain the overall memory size.
The method according to claim 3, wherein, using each training process in the plurality of training processes, calculating the memory size required to support caching a group of images read in the plurality of groups of images, and obtaining the same as The multiple memory sizes corresponding to the multiple groups of images one-to-one include:

Using each training process in the plurality of training processes to obtain shape information of a group of images read in the plurality of groups of images;

Using each training process in the plurality of training processes, according to the shape information of a group of images read in the plurality of groups of images, calculate the memory size required by a group of images that support cache reading, and obtain the described Multiple memory sizes.
The method according to claim 5, wherein, after obtaining the shape information of a group of images read in the plurality of groups of images by using each training process in the plurality of training processes, the method further include:

Using each training process in the multiple training processes to summarize the shape information of the multiple groups of images to obtain an information summary result.
An image caching device, comprising:

The reading module is configured to use each training process in multiple training processes to read a group of images to obtain multiple groups of images; wherein, the multiple training processes correspond to the multiple groups of images one-to-one;

The processing module is configured to use the first training process in the plurality of training processes to apply for the shared memory corresponding to the plurality of groups of images, and share the applied for shared memory to the plurality of training processes and all each training session different from the first training session;

The caching module is configured to use each training process in the plurality of training processes to respectively cache a group of images read in the plurality of groups of images into the shared memory for the plurality of training processes Each training process reads the multiple sets of images from the shared memory during the execution of the neural network training step.
The apparatus according to claim 7, wherein,

The reading module is specifically configured to use the second training process in the plurality of training processes to read the image path list that records the storage path of each image in the image data set, and broadcast the image path list to the Each training process different from the second training process in the multiple training processes; using each training process in the multiple training processes, based on the image path list, according to the corresponding preset image reading strategy, from A group of images is read from the image data set to obtain the multiple groups of images.
The apparatus according to claim 7, wherein,

The processing module is further configured to use each training process in the plurality of training processes to calculate the memory size required to support the caching of a group of images read in the plurality of groups of images, and obtain the memory size corresponding to the plurality of groups of images Multiple memory sizes corresponding to each other; using the first training process, summing up the multiple memory sizes to obtain an overall memory size that supports storing the multiple groups of images;

The processing module is specifically configured to use the first training process to apply for the shared memory according to the overall memory size.
The apparatus according to claim 9, wherein,

The processing module is further configured to use each training process in the multiple training processes to summarize the multiple memory sizes to obtain the overall memory size.
The apparatus according to claim 9, wherein,

The processing module is specifically configured to use each training process in the plurality of training processes to acquire the shape information of a group of images read in the plurality of groups of images; use each of the plurality of training processes to The training process, according to the shape information of a group of images read in the plurality of groups of images, calculates the memory size required by a group of images that support cache reading, and obtains the multiple memory sizes.
The apparatus of claim 11, wherein,

The processing module is further configured to use each training process in the multiple training processes to summarize the shape information of the multiple groups of images to obtain an information summary result.
A kind of electronic equipment, described electronic equipment comprises: processor, memory and communication bus; Wherein,

The communication bus is configured to realize connection and communication between the processor and the memory;

The processor is configured to execute one or more programs stored in the memory, so as to realize the image caching method according to any one of claims 1-6.
A computer-readable storage medium, the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors, so as to realize any one of claims 1-6 The image caching method described.
A computer program product, the computer program product comprising a computer program or instruction, when the computer program or instruction is run on a computer, the computer is made to execute the image cache described in any one of claims 1-6 method.