CN116011550A - A model pruning method, image processing method and related device - Google Patents

Abstract

The invention provides a model pruning method, an image processing method and a related device, comprising the following steps: the method comprises the steps of obtaining a model to be pruned and a corresponding pruning target thereof, wherein the model to be pruned is obtained based on image data training, the pruning target comprises pruning proportion information, and the pruning proportion information is used for representing proportion information of primary pruning and iterative pruning in a model pruning process; pruning is carried out on the model to be pruned for one time according to the pruning target, and a first pruned model is obtained; and carrying out iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and taking the second pruned model as a target model. The invention can accelerate the training process, and compared with a single pruning method, the pruning upper limit is improved.

Description

A model pruning method, image processing method and related device

technical field

The invention relates to the technical field of model compression, in particular to a model pruning method, an image processing method and related devices.

Background technique

With the development of artificial intelligence technology, the application of neural network models is becoming more and more extensive. Considering that the network model has many parameters and a large amount of calculation, in order to improve the calculation speed of the model, it is necessary to compress the model. Through model compression, the purpose of reducing model size, reducing resource consumption and improving response time is achieved.

Among them, model pruning is a common model compression method at present. This method achieves compression of the neural network model by removing parameters with smaller weights in the network model after the model is trained. The current model pruning is divided into iterative pruning and one-shot pruning. Among them, iterative pruning requires fine-tune training after each pruning in the pruning training process. This leads to more time-consuming training. Therefore, how to speed up the training process of iterative pruning in iterative pruning is a technical problem that needs to be solved urgently.

Contents of the invention

The present invention provides a model pruning method, an image processing method and related devices to solve the above problems.

The present invention provides a method for model pruning, comprising:

Obtaining the model to be pruned and its corresponding pruning target; wherein, the model to be pruned is obtained based on image data training, the pruning target includes pruning ratio information, and the pruning ratio information is used to represent the Information on the proportion of one-time pruning and iterative pruning in the model pruning process;

Performing a pruning on the model to be pruned according to the pruning target to obtain a first pruned model;

Perform iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and use the second pruned model as a target model.

According to a model pruning method provided by the present invention, the pruning target includes the total number of pruning training iterations;

Correspondingly, after performing a pruning on the model to be pruned according to the pruning target to obtain the first pruned model, the method further includes:

Perform fine-tuning training on the first pruned model according to the first number of fine-tuning times to obtain the first fine-tuned model; wherein, the first number of fine-tuning times is based on the first pruning in the pruning ratio information in the model pruning process The proportion information in and the total number of pruning training iterations are determined;

The iterative pruning of the first pruned model according to the pruning target to obtain the second pruned model includes:

Iteratively pruning the first fine-tuned model according to the pruning target to obtain a second pruned model.

According to a model pruning method provided by the present invention, the pruning target further includes the number of target pruning channels;

Correspondingly, performing iterative pruning on the first fine-tuned model according to the pruning target to obtain a second pruned model includes:

S1, pruning a channel in the first fine-tuned model to obtain a single-channel pruned model;

S2. Perform fine-tuning training on the single-channel pruned model according to the second fine-tuning times to obtain the second fine-tuned model; wherein, the second fine-tuning times are based on the proportion of iterative pruning in the model pruning process in the pruning ratio information The ratio information, the number of target pruning channels and the total number of pruning training iterations are determined;

S3. Repeat S1-S2 on the basis of the second fine-tuned model until the preset number of iterations is reached, so as to obtain the latest second fine-tuned model, and use the latest second fine-tuned model as the second trimmed Branch model;

Wherein, the preset number of iterations is determined according to the number of iterative pruning channels, and the number of iterative pruning channels is based on the number of target pruning channels and the number of iterative pruning in the model pruning process in the pruning ratio information. The proportion information is calculated.

According to a model pruning method provided by the present invention, the second fine-tuning times are determined according to the proportion information of iterative pruning in the model pruning process in the pruning ratio information, the target number of pruning channels, and the total number of pruning training iterations ,include:

According to the proportion information of iterative pruning in the model pruning process and the total number of pruning training iterations, the total number of iterative pruning fine-tuning training is calculated;

The second fine-tuning times are calculated according to the total times of iterative pruning and fine-tuning training and the number of iterative pruning channels.

According to a model pruning method provided by the present invention, the pruning target further includes the number of target pruning channels;

Correspondingly, performing a pruning on the model to be pruned according to the pruning target to obtain a first pruned model, including:

Obtain the first norm of each channel in the model to be pruned, and sort the channels in the model to be pruned according to the value of the first norm from small to large, to obtain a channel sequence;

Taking the first p channels in the channel sequence as the first channel to be pruned, where p is based on the number of target pruning channels and the proportion of one pruning in the model pruning process in the pruning ratio information The information is calculated;

Pruning the model to be pruned according to the first channel to be pruned to obtain a first pruned model.

According to a model pruning method provided by the present invention, said S1 prunes a channel in the first fine-tuned model to obtain a single-channel pruned model, including:

Obtain the second norm of each channel in the first fine-tuned model;

Use the channel with the smallest value of the second norm as the second channel to be pruned;

Pruning the first fine-tuned model according to the second channel to be pruned to obtain a single-channel pruned model.

The present invention also provides an image processing method, comprising:

Get the image to be processed;

Inputting the image to be processed into the trained image processing model, processing the image to be processed by the trained image processing model to obtain a processing result;

Wherein, the trained image processing model is obtained by any of the above-mentioned model pruning methods.

The present invention also provides a model pruning device, comprising:

The model and pruning target acquisition module is used to obtain the model to be pruned and its corresponding pruning target; wherein, the model to be pruned is obtained based on image data training, and the pruning target includes pruning ratio information, so The above pruning ratio information is used to represent the proportion information of one-time pruning and iterative pruning in the model pruning process;

A pruning module, configured to perform one pruning on the model to be pruned according to the pruning target to obtain the first pruned model;

An iterative pruning module, configured to perform iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and use the second pruned model as a target model.

The present invention also provides an image processing device, comprising:

An image acquisition module, configured to acquire images to be processed;

An image processing module, configured to input the image to be processed into a trained image processing model, and process the image to be processed through the trained image processing model to obtain a processing result; wherein, the trained The image processing model of is obtained by the model pruning device as described above.

The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and operable on the processor. When the processor executes the program, any one of the above-mentioned model pruning methods or image processing methods is implemented.

The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, any one of the above-mentioned model pruning methods or image processing methods is implemented.

The model pruning method, image processing method and related devices provided by the present invention, wherein, the model pruning method can accelerate the training process through the two methods of one-time pruning and iterative pruning in the present invention, and compared with a single For the one-time pruning method, the upper limit of pruning is increased.

Description of drawings

In order to more clearly illustrate the present invention or the technical solutions in the prior art, the accompanying drawings that need to be used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings in the following description are the present invention. For some embodiments of the invention, those skilled in the art can also obtain other drawings based on these drawings without creative effort.

Fig. 1 is one of the schematic flow charts of the model pruning method provided by the embodiment of the present invention;

Fig. 2 is the second schematic flow diagram of the model pruning method provided by the embodiment of the present invention;

FIG. 3 is a schematic flowchart of an image processing method provided by an embodiment of the present invention;

Fig. 4 is the structural representation of the model pruning device provided by the embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an image processing device provided by an embodiment of the present invention;

FIG. 6 is a schematic diagram of a physical structure of an electronic device provided by an embodiment of the present invention.

Detailed ways

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

Fig. 1 is one of the schematic flow diagrams of the model pruning method provided by the embodiment of the present invention; Fig. 2 is the second schematic flow diagram of the model pruning method provided by the embodiment of the present invention; as shown in Fig. 1 and Fig. 2, the model pruning Branch methods, including:

S101. Obtain a model to be pruned and its corresponding pruning target.

Wherein, the model to be pruned is obtained based on image data training, and the pruning target includes pruning ratio information, and the pruning ratio information is used to represent the results of one-time pruning and iterative pruning in the model pruning process. percentage information.

In this step, the model to be pruned is a pre-training model, and the pre-training model may specifically be a convolutional neural network, which is trained based on training images and corresponding image labels, and is used for image processing.

The pruning target can be the size of the pre-trained model, the pruning rate of the entire pre-trained model, or the number of pruning channels, or the total number of iterations of pruning training. In the present invention, the pruning target may also include pruning ratio information used to characterize the proportion of primary pruning and iterative pruning in the model pruning process, which is used to configure two types of primary pruning and iterative pruning. The proportion of each pruning method in the whole pruning training process.

S102. Perform one pruning on the model to be pruned according to the pruning target to obtain a first pruned model.

In this step, the one-shot pruning method is used to prune the model to be pruned (one-shot pruning) according to the pruning target (such as pruning ratio information), and the pruned model is used as the first pruning method. Branch model.

S103. Perform iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and use the second pruned model as a target model.

In this step, based on the above-mentioned first pruned model, pruning is continued according to the pruning target, the iterative pruning method is used to prune the first pruned model, and the pruned model is used as the second A pruned model, the second pruned model is a final output target model for image processing.

In the prior art, fine-tune training (fine-tune training) is required after each iterative pruning. Assuming that the model to be pruned is iteratively pruned, and the number of iterations is 10, fine-tune training after each iterative pruning If the number of times is 1000, the total number of fine-tuning training is 10000; and if the model to be pruned is pruned once, and the number of fine-tuning training is set to 2000, the total number of fine-tuning training is 2000. From the above examples, it can be seen that the total number of fine-tuning training in iterative pruning is too much compared with the one-time pruning method, resulting in a long time-consuming pruning training process. The upper limit of the branch model will be limited. Therefore, the present invention combines two methods of one-time pruning and iterative pruning. For example, one-time pruning is performed first. In one pruning process, the number of times of fine-tuning training is 1000. The number of times is 5, the number of times of fine-tuning training is 1000, then the total number of times of fine-tuning training in the whole pruning process is 1000+5*1000=6000. By comparing 6000 and 10000, it can be seen that by combining the two pruning methods of one-time pruning and iterative pruning, the number of fine-tuning training can be greatly reduced, thereby shortening the training time. In addition, compared with simply using the one-time pruning method, the upper limit of pruning can also be increased.

The model pruning method provided by the embodiment of the present invention can speed up the training process through the two methods of one-time pruning and iterative pruning in the present invention, and compared with the single-time pruning method, the pruning method is improved. upper limit.

In some embodiments of the present invention, the pruning target includes pruning ratio information and the total number of pruning training iterations iters_all, and the pruning ratio information is used to represent the difference between one pruning and iterative pruning Branch proportion information, the pruning ratio information is more specifically the pruning ratio value, assuming that the proportion of one pruning is α(α∈[0,1]), then the proportion of iterative pruning is 1- alpha.

Correspondingly, after performing a pruning on the model to be pruned according to the pruning target to obtain the first pruned model, the method further includes:

Perform fine-tuning training on the first pruned model according to the first number of fine-tuning times to obtain the first fine-tuned model.

Wherein, the first fine-tuning times are determined according to the proportion information of one pruning in the model pruning process in the pruning ratio information and the total number of pruning training iterations iters_all. Specifically, if the proportion of one pruning is α, the first fine-tuning times iter1=0.8*α*iters_all, where 0.8 is a set constant.

The iterative pruning of the first pruned model according to the pruning target to obtain the second pruned model includes:

Iteratively pruning the first fine-tuned model according to the pruning target to obtain a second pruned model. That is, iterative pruning is performed on the basis of one-shot pruning.

In addition, in order to further accelerate the pruning training process, the proportion α of one-time pruning is set to 0.8 according to the empirical value, and the corresponding proportion of iterative pruning is 0.2, so as to speed up the pruning to the greatest extent while ensuring the pruning effect training process.

The model pruning method provided by the embodiment of the present invention speeds up the pruning training process by determining the number of fine-tuning training times for the model after the first pruning according to the proportion information of one pruning in the model pruning process .

In some embodiments of the present invention, the pruning target further includes a target number of pruning channels c_target.

Correspondingly, performing iterative pruning on the first fine-tuned model according to the pruning target to obtain a second pruned model includes:

S1. Pruning one channel in the first fine-tuned model to obtain a single-channel pruned model.

S2. Perform fine-tuning training on the single-channel pruned model according to the second number of fine-tuning times to obtain a second fine-tuned model.

Wherein, the second fine-tuning times are determined according to the proportion information of iterative pruning in the model pruning process in the pruning ratio information, the target number of pruning channels, and the total number of pruning training iterations.

Specifically, the second number of fine-tuning times is determined according to the proportion information of iterative pruning in the model pruning process in the pruning ratio information, the number of target pruning channels, and the total number of pruning training iterations, including:

According to the proportion information of iterative pruning in the model pruning process (the proportion of iterative pruning is 1-α) and the total number of pruning training iterations iters_all is calculated to obtain the total number of iterative pruning fine-tuning training = (1-0.8*α) *iters_all.

The number of times of fine-tuning to be performed by the model in each iteration (ie, the second number of times of fine-tuning iter2) is calculated according to the total number of training times of iterative pruning and fine-tuning and the number of iterative pruning channels c2. specifically,

S3. Repeat S1-S2 on the basis of the second fine-tuned model until the preset number of iterations is reached, so as to obtain the latest second fine-tuned model, and use the latest second fine-tuned model as the second trimmed Branch model. That is, one channel in the second fine-tuned model is pruned to obtain a new single-channel pruned model. Perform fine-tuning training on the new single-channel pruned model according to the second number of fine-tuning times to obtain a new second fine-tuned model. This cycle is repeated until the preset number of iterations is reached (that is, the pruning is completed according to the target number of pruning channels).

Wherein, the preset number of iterations iter2 is determined according to the number of iterative pruning channels c2, and the number of iterative pruning channels c2 is based on the number of iterative pruning channels in the target pruning channel The proportion information (1-α) in the branch process is calculated, specifically, c2=c_target*(1-α).

The model pruning method provided by the embodiment of the present invention performs iterative pruning on the basis of one pruning, which not only speeds up the pruning training process, but also increases the upper limit of pruning.

In some embodiments of the present invention, the pruning target further includes a target number of pruning channels c_target.

Correspondingly, performing a pruning on the model to be pruned according to the pruning target to obtain a first pruned model, including:

Obtain the first norm of each channel in the model to be pruned, and sort the channels in the model to be pruned according to the value of the first norm from small to large, to obtain a channel sequence.

Specifically, calculate the sum of the norm values of the weights in each channel under each layer of the network layer in the model to be pruned, and according to the sum of the norm values of the weights corresponding to each channel (that is, the first norm) according to the order from small to large Each channel is sorted sequentially, and the total number of channels is recorded as T.

The first p channels in the channel sequence are used as the first channels to be pruned.

Wherein, p is calculated according to the target number of pruning channels c_target and the proportion information α of one pruning in the model pruning process in the pruning ratio information. Specifically, p=c_target*α.

Pruning the model to be pruned according to the first channel to be pruned to obtain a first pruned model. That is, p p channels are pruned in the model to be pruned according to the identification information of the first channel to be pruned to obtain the first pruned model.

In this embodiment, the norm value of the weight is obtained by performing L1 norm calculation on the weight, and in other solutions of the present invention, it may also be calculated by other means such as L0 norm and L2 norm.

In the model pruning method provided by the embodiment of the present invention, the channel with lower pruning sensitivity is determined as the first channel to be pruned according to the norm value of each channel, and one pruning is performed to obtain the first pruned model.

In some embodiments of the present invention, the S1 is to prune one channel in the first fine-tuned model to obtain a single-channel pruned model, including:

Obtain the second norm of each channel in the first fine-tuned model. That is, the sum of the norm values of the channel weights in all network layers in the first fine-tuned model is calculated.

The channel with the smallest value of the second norm is used as the second channel to be pruned.

Pruning the first fine-tuned model according to the second channel to be pruned to obtain a single-channel pruned model. That is, in the iterative pruning process, only one channel is pruned in each iteration.

In this embodiment, the norm value of the weight is obtained by performing L1 norm calculation on the weight, and in other solutions of the present invention, it may also be calculated by other means such as L0 norm and L2 norm.

Fig. 3 is a schematic flow chart of an image processing method provided by an embodiment of the present invention; as shown in Fig. 3, the image processing method includes:

S301. Acquire an image to be processed.

S302. Input the image to be processed into a trained image processing model, and process the image to be processed through the trained image processing model to obtain a processing result.

Wherein, the trained image processing model is obtained through the above-mentioned model pruning method.

In this embodiment, the trained image processing model is obtained through the above model pruning method, and the trained image processing model is transplanted into a terminal with limited computing power. After the terminal acquires the image to be processed, it will The image is input to the trained image processing model, and the prediction result is obtained directly, reducing resource consumption and improving response time. For example, the image processing model is a signboard classification model. After training and pruning on the cloud or other devices with rich computing resources, it is transplanted to the autonomous driving vehicle. After the vehicle terminal acquires the images around the vehicle, it uses the signboard classification model to perform classification prediction and obtain the classification result. Or, the image processing model is a face attribute recognition model, which is used to identify the gender, age, race, etc. of people in the input face image. Then, the face attribute recognition model is transplanted to computing resources after training and pruning In the limited terminal, the face image acquired by the terminal is recognized to obtain a face recognition result.

The image processing method provided by the embodiment of the present invention obtains an image processing model based on the pruning method mentioned above, and uses the image processing model to realize image processing in a terminal with limited computing resources, thereby achieving the purpose of reducing resource consumption and improving response time .

The model pruning device and image processing device provided by the present invention are described below. The model pruning device described below and the model pruning method described above can be referred to in correspondence with each other, and the image processing device and image processing method can be referred to in correspondence.

FIG. 4 is a schematic structural diagram of a model pruning device provided by an embodiment of the present invention. As shown in FIG. 4 , the model pruning device includes a model and pruning target acquisition module 401 , a primary pruning module 402 and an iterative pruning module 403 .

The model and pruning target acquisition module 401 is used to acquire the model to be pruned and its corresponding pruning target.

Wherein, the model to be pruned is obtained based on image data training, and the pruning target includes pruning ratio information, and the pruning ratio information is used to represent the results of one-time pruning and iterative pruning in the model pruning process. percentage information.

In this module, the model to be pruned is a pre-training model, and the pre-training model can be specifically a convolutional neural network, and it is trained based on training images and corresponding image labels for image processing.

The pruning target can be the size of the pre-trained model, the pruning rate of the entire pre-trained model, or the number of pruning channels, or the total number of iterations of pruning training. In the present invention, the pruning target may also include pruning ratio information used to characterize the proportion of primary pruning and iterative pruning in the model pruning process, which is used to configure two types of primary pruning and iterative pruning. The proportion of each pruning method in the whole pruning training process.

A pruning module 402, configured to perform one pruning on the model to be pruned according to the pruning target to obtain a first pruned model.

In this module, the one-shot pruning method is used to prune the model to be pruned (one-shot pruning) according to the pruning target (such as pruning ratio information), and the pruned model is used as the first pruning model. Branch model.

The iterative pruning module 403 is configured to perform iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and use the second pruned model as a target model.

In this module, based on the first pruned model mentioned above, pruning is continued according to the pruning target, the iterative pruning method is used to prune the first pruned model, and the pruned model is used as the second A pruned model, the second pruned model is a final output target model for image processing.

In the prior art, fine-tune training (fine-tune training) is required after each iterative pruning. Assuming that the model to be pruned is iteratively pruned, and the number of iterations is 10, fine-tune training after each iterative pruning If the number of times is 1000, the total number of fine-tuning training is 10000; and if the model to be pruned is pruned once, and the number of fine-tuning training is set to 2000, the total number of fine-tuning training is 2000. From the above examples, it can be seen that the total number of fine-tuning training in iterative pruning is too much compared with the one-time pruning method, resulting in a long time-consuming pruning training process. The upper limit of the branch model will be limited. Therefore, the present invention combines two methods of one-time pruning and iterative pruning. For example, one-time pruning is performed first. In one pruning process, the number of times of fine-tuning training is 1000. The number of times is 5, the number of times of fine-tuning training is 1000, then the total number of times of fine-tuning training in the whole pruning process is 1000+5*1000=6000. By comparing 6000 and 10000, it can be seen that by combining the two pruning methods of one-time pruning and iterative pruning, the number of fine-tuning training can be greatly reduced, thereby shortening the training time. In addition, compared with simply using the one-time pruning method, the upper limit of pruning can also be increased.

The model pruning device provided by the embodiment of the present invention can accelerate the training process through the two methods of one-time pruning and iterative pruning in the present invention, and compared with the single-time pruning method, the pruning method is improved. upper limit.

FIG. 5 is a schematic structural diagram of an image processing device provided by an embodiment of the present invention; as shown in FIG. 5 , the image processing device includes an image acquisition module 501 and an image processing module 502 .

An image acquisition module 501, configured to acquire an image to be processed.

The image processing module 502 is configured to input the image to be processed into a trained image processing model, and process the image to be processed through the trained image processing model to obtain a processing result.

Wherein, the trained image processing model is obtained through the above-mentioned model pruning device.

In this embodiment, the trained image processing model is obtained through the above model pruning method, and the trained image processing model is transplanted into a terminal with limited computing power. After the terminal acquires the image to be processed, it will The image is input to the trained image processing model, and the prediction result is obtained directly, reducing resource consumption and improving response time. For example, the image processing model is a signboard classification model. After training and pruning on the cloud or other devices with rich computing resources, it is transplanted to the autonomous driving vehicle. After the vehicle terminal acquires the images around the vehicle, it uses the signboard classification model to perform classification prediction and obtain the classification result. Or, the image processing model is a face attribute recognition model, which is used to identify the gender, age, race, etc. of people in the input face image. Then, the face attribute recognition model is transplanted to computing resources after training and pruning In the limited terminal, the face image acquired by the terminal is recognized to obtain a face recognition result.

The image processing device provided by the embodiment of the present invention obtains an image processing model based on the above-mentioned pruning method, and uses the image processing model to realize image processing in a terminal with limited computing resources, thereby achieving the purpose of reducing resource consumption and improving response time .

FIG. 6 is a schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. As shown in FIG. 6, the electronic device may include: a processor (processor) 610, a communication interface (Communications Interface) 620, and a memory (memory) 630 and a communication bus 640 , wherein the processor 610 , the communication interface 620 and the memory 630 complete mutual communication through the communication bus 640 . The processor 610 can call the logic instructions in the memory 630 to execute the model pruning method, the model pruning method includes: obtaining the model to be pruned and its corresponding pruning target, wherein the model to be pruned is based on Obtained from image data training, the pruning target includes pruning ratio information, and the pruning ratio information is used to represent the proportion information of one-time pruning and iterative pruning in the model pruning process; according to the pruning The target is to perform one pruning on the model to be pruned to obtain a first pruned model; to perform iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and The second pruned model is used as the target model.

Or to execute an image processing method, the image processing method includes: acquiring an image to be processed; inputting the image to be processed into a trained image processing model, and processing the image to be processed by the trained image processing model The image is processed to obtain a processing result; wherein, the trained image processing model is obtained through the above-mentioned model pruning method.

In addition, the logic instructions in the above-mentioned memory 630 may be implemented in the form of software functional units and when sold or used as an independent product, may be stored in a computer-readable storage medium. Based on this understanding, the essence of the technical solution of the present invention or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including Several instructions are used to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods in various embodiments of the present invention. The aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disk or optical disc, etc., which can store program codes. .

In yet another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it is implemented to perform a model pruning method, and the model pruning method includes : Acquiring the model to be pruned and its corresponding pruning target, wherein the model to be pruned is trained based on image data, the pruning target includes pruning ratio information, and the pruning ratio information is used to represent In the process of model pruning, the proportion information of one-time pruning and iterative pruning; perform one-time pruning on the model to be pruned according to the pruning target to obtain the first pruned model; according to the pruning target Perform iterative pruning on the first pruned model to obtain a second pruned model, and use the second pruned model as a target model.

Or to execute an image processing method, the image processing method includes: acquiring an image to be processed; inputting the image to be processed into a trained image processing model, and processing the image to be processed by the trained image processing model The image is processed to obtain a processing result; wherein, the trained image processing model is obtained through the above-mentioned model pruning method.

The device embodiments described above are only illustrative, and the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place , or can also be distributed to multiple network elements. Part or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. It can be understood and implemented by those skilled in the art without any creative efforts.

Through the above description of the implementations, those skilled in the art can clearly understand that each implementation can be implemented by means of software plus a necessary general hardware platform, and of course also by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of software products, and the computer software products can be stored in computer-readable storage media, such as ROM/RAM, magnetic Discs, optical discs, etc., include several instructions to make a computer device (which may be a personal computer, server, or network device, etc.) execute the methods of various embodiments or some parts of the embodiments.

Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: it can still be Modifications are made to the technical solutions described in the foregoing embodiments, or equivalent replacements are made to some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims (11)

1. A method of pruning a model, comprising:

obtaining a model to be pruned and a corresponding pruning target; the pruning target comprises pruning proportion information which is used for representing proportion information of primary pruning and iterative pruning in the pruning process of the model;

pruning is carried out on the model to be pruned for one time according to the pruning target, and a first pruned model is obtained;

and carrying out iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and taking the second pruned model as a target model.

2. The model pruning method of claim 1, wherein the pruning target further comprises a total number of pruning training iterations;

correspondingly, after pruning the model to be pruned for one time according to the pruning target to obtain a first pruned model, the method further comprises:

performing fine tuning training on the first pruned model according to the first fine tuning times to obtain a first fine tuned model; the first fine tuning times are determined according to the proportion information of one pruning in the pruning proportion information in the model pruning process and the total number of pruning training iterations;

performing iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, including:

and carrying out iterative pruning on the first fine-tuned model according to the pruning target to obtain a second pruned model.

3. The model pruning method according to claim 2, wherein the pruning target further comprises a target pruning channel number;

correspondingly, the iterative pruning is carried out on the first fine-tuned model according to the pruning target to obtain a second pruned model, which comprises the following steps:

s1, pruning is carried out on one channel in the first fine-tuned model, and a single-channel pruned model is obtained;

s2, performing fine tuning training on the single-channel pruned model according to the second fine tuning times to obtain a second fine tuned model; the second fine tuning times are determined according to the proportion information of iterative pruning in the model pruning process, the target pruning channel number and the pruning training iteration total number in the pruning proportion information;

s3, repeatedly executing S1-S2 on the basis of the second fine-tuned model until the preset iteration times are reached, so as to obtain a latest second fine-tuned model, and taking the latest second fine-tuned model as a second pruned model;

the preset iteration times are determined according to the number of iteration pruning channels, and the number of iteration pruning channels is calculated according to the number of target pruning channels and the duty ratio information of the iteration pruning in the model pruning process in the pruning proportion information.

4. The method of model pruning according to claim 3, wherein the second fine tuning number is determined according to the ratio information of iterative pruning in the model pruning process, the target pruning channel number and the total number of pruning training iterations in the pruning proportion information, and includes:

calculating to obtain the total number of iterative pruning fine tuning training according to the duty ratio information of the iterative pruning in the model pruning process and the total number of pruning training iterations;

and calculating the second fine tuning times according to the total times of the iterative pruning fine tuning training and the number of the iterative pruning channels.

5. The model pruning method according to claim 2, wherein the pruning target further comprises a target pruning channel number;

correspondingly, the pruning is carried out on the model to be pruned for one time according to the pruning target to obtain a first pruned model, which comprises the following steps:

acquiring a first norm of each channel in the model to be pruned, and sequencing each channel in the model to be pruned according to the sequence from small to large of the numerical value of the first norm to obtain a channel sequence;

taking the first p channels in the channel sequence as first channels to be pruned, wherein p is calculated according to the number of the target pruning channels and the duty ratio information of primary pruning in the pruning proportion information in the model pruning process;

pruning is carried out on the to-be-pruned model according to the first to-be-pruned channel, and a first pruned model is obtained.

6. The method for pruning a model according to claim 3, wherein said step S1 of pruning one channel in said first trimmed model to obtain a single channel pruned model comprises:

acquiring a second norm of each channel in the first fine-tuned model;

taking the channel with the smallest second norm value as a second channel to be pruned;

pruning the first fine-tuned model according to the second to-be-pruned channel to obtain a single-channel pruned model.

7. An image processing method, comprising:

acquiring an image to be processed;

inputting the image to be processed into a trained image processing model, and processing the image to be processed through the trained image processing model to obtain a processing result;

wherein the trained image processing model is obtained by the model pruning method according to any one of claims 1-6.

8. A model pruning device, comprising:

the model and pruning target acquisition module is used for acquiring a model to be pruned and a pruning target corresponding to the model to be pruned; the pruning target comprises pruning proportion information which is used for representing proportion information of primary pruning and iterative pruning in the pruning process of the model;

the primary pruning module is used for pruning the model to be pruned for one time according to the pruning target to obtain a first pruned model;

and the iterative pruning module is used for carrying out iterative pruning on the first pruned model according to the pruning target to obtain a second pruned model, and taking the second pruned model as a target model.

9. An image processing apparatus, comprising:

the image acquisition module is used for acquiring an image to be processed;

the image processing module is used for inputting the image to be processed into a trained image processing model, and processing the image to be processed through the trained image processing model to obtain a processing result; wherein the trained image processing model is obtained by the model pruning device according to claim 8.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the model pruning method of any one of claims 1 to 6 or the image processing method of claim 7 when executing the program.

11. A non-transitory computer readable storage medium having stored thereon a computer program, characterized in that the computer program, when executed by a processor, implements the model pruning method of any one of claims 1 to 6 or the image processing method of claim 7.

Images