CN113506229A

CN113506229A - Neural network training and image generation method and device

Info

Publication number: CN113506229A
Application number: CN202110799904.5A
Authority: CN
Inventors: 施路平; 杨哲宇; 赵蓉
Original assignee: Tsinghua University
Current assignee: Tsinghua University
Priority date: 2021-07-15
Filing date: 2021-07-15
Publication date: 2021-10-15
Anticipated expiration: 2041-07-15
Also published as: CN113506229B

Abstract

The present disclosure relates to a neural network training and image generating method and apparatus, the method comprising: inputting the dynamic visual information of the sample scene into a first reconstruction network for processing to obtain a first reconstruction result; inputting the sample color image of the sample scene into a second reconstruction network for processing to obtain a second reconstruction result; determining the comprehensive network loss of the first reconstruction network and the second reconstruction network according to the sample color image, the first reconstruction result and the second reconstruction result; training a first reconstruction network and the second reconstruction network based on the synthetic network loss. According to the neural network training method disclosed by the embodiment of the disclosure, the feature graph corresponding to the dynamic visual information is closer to the feature graph corresponding to the color image through training, the second reconstructed image output by the second reconstruction network can be closer to the real color image, and the accuracy and the fidelity of the first reconstructed image close to the second reconstructed image can be improved.

Description

Neural network training and image generation method and device

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to a method and an apparatus for neural network training and image generation.

Background

In the related art, the frame frequency of images or video frames acquired by a camera or a video camera is not high, the number of the acquired video frames in a certain period of time is limited, and if the speed of a shot target object is high, the target object is difficult to shoot and the pose of the target object is difficult to determine in the time interval between two video frames, so that the action or the track of the target object is missed.

Disclosure of Invention

The disclosure provides a neural network training and image generating method and device.

According to an aspect of the present disclosure, there is provided an image neural network training method, including: inputting sample dynamic visual information of a sample scene into a first reconstruction network for processing to obtain a first reconstruction result, wherein the first reconstruction result comprises a multi-stage first feature map and a first reconstruction image; inputting a sample color image of the sample scene into a second reconstruction network for processing to obtain a second reconstruction result, wherein the second reconstruction result comprises a multi-level second feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time; determining the comprehensive network loss of the first reconstruction network and the second reconstruction network according to the sample color image, the first reconstruction result and the second reconstruction result; and training the first reconstruction network and the second reconstruction network according to the comprehensive network loss, wherein the first reconstruction network is used for generating color images according to dynamic visual information, and the second reconstruction network is used for training the first reconstruction network.

In one possible implementation, determining a composite network loss of the first reconstruction network and the second reconstruction network according to the sample color image, the first reconstruction result, and the second reconstruction result includes: determining a first network loss according to the first reconstruction result and the second reconstruction result; determining a second network loss from the second reconstructed image and the sample color image; and determining the comprehensive network loss according to the first network loss and the second network loss.

In one possible implementation, determining a first network loss according to the first reconstruction result and the second reconstruction result includes: respectively determining first sub-losses of each stage according to the first characteristic diagram of each stage and a second characteristic diagram with the same resolution as the first characteristic diagram of each stage; determining a second sub-loss according to the first reconstructed image and the second reconstructed image; and carrying out weighted summation processing on the first sub-losses and the second sub-losses of each stage to obtain the first network loss.

In one possible implementation, determining the composite network loss according to the first network loss and the second network loss includes: and carrying out weighted summation processing on the first network loss and the second network loss to obtain the comprehensive network loss.

In one possible implementation, the first reconstruction network comprises a recurrent neural network and the second reconstruction network comprises a convolutional neural network.

According to an aspect of the present disclosure, there is provided an image generation method including: and inputting the dynamic visual information of the preset scene acquired at a plurality of moments in a first time period into the first reconstruction network trained by the neural network training method for processing to generate a first color image corresponding to each dynamic visual information.

In one possible implementation, the method further includes: and obtaining a video of the preset scene in the first time period according to the first color image and a second color image of the preset scene obtained in the first time period, wherein the obtaining frequency of the dynamic visual information is higher than that of the color image.

According to an aspect of the present disclosure, there is provided a neural network training apparatus including: the first reconstruction module is used for inputting the sample dynamic visual information of the sample scene into a first reconstruction network for processing to obtain a first reconstruction result, and the first reconstruction result comprises a multi-stage first feature map and a first reconstruction image; the second reconstruction module is used for inputting a sample color image of the sample scene into a second reconstruction network for processing to obtain a second reconstruction result, the second reconstruction result comprises a second multi-level feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time; a loss determining module, configured to determine, according to the sample color image, the first reconstruction result, and the second reconstruction result, a comprehensive network loss of the first reconstruction network and the second reconstruction network; and the training module is used for training the first reconstruction network and the second reconstruction network according to the comprehensive network loss, wherein the first reconstruction network is used for generating a color image according to dynamic visual information, and the second reconstruction network is used for training the first reconstruction network.

In one possible implementation, the loss determination module is further configured to: determining a first network loss according to the first reconstruction result and the second reconstruction result; determining a second network loss from the second reconstructed image and the sample color image; and determining the comprehensive network loss according to the first network loss and the second network loss.

In one possible implementation, the loss determination module is further configured to: respectively determining first sub-losses of each stage according to the first characteristic diagram of each stage and a second characteristic diagram with the same resolution as the first characteristic diagram of each stage; determining a second sub-loss according to the first reconstructed image and the second reconstructed image; and carrying out weighted summation processing on the first sub-losses and the second sub-losses of each stage to obtain the first network loss.

In a possible implementation manner, the first network loss and the second network loss are subjected to weighted summation processing, so as to obtain the comprehensive network loss.

According to an aspect of the present disclosure, there is provided an image generation apparatus including: and the generating module is used for inputting the dynamic visual information of the preset scene acquired at a plurality of moments in a first time period into the first reconstruction network trained by the neural network training device for processing, and generating a first color image corresponding to each dynamic visual information.

In one possible implementation, the apparatus further includes: and the video generation module is used for obtaining a video in the first time period of the preset scene according to the first color image and the second color image of the preset scene obtained in the first time period, wherein the obtaining frequency of the dynamic visual information is higher than that of the color image.

According to an aspect of the present disclosure, there is provided an electronic device including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the memory-stored instructions to perform the above-described method.

According to an aspect of the present disclosure, there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, implement the above-described method.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure. Other features and aspects of the present disclosure will become apparent from the following detailed description of exemplary embodiments, which proceeds with reference to the accompanying drawings.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the disclosure.

FIG. 1 shows a flow diagram of a neural network training method in accordance with an embodiment of the present disclosure;

FIG. 2 illustrates an application diagram of a neural network training method in accordance with an embodiment of the present disclosure;

FIG. 3 shows a block diagram of a neural network training device, in accordance with an embodiment of the present disclosure;

FIG. 4 shows a block diagram of an electronic device according to an embodiment of the disclosure;

fig. 5 illustrates a block diagram of an electronic device according to an embodiment of the disclosure.

Detailed Description

Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. In the drawings, like reference numbers can indicate functionally identical or similar elements. While the various aspects of the embodiments are presented in drawings, the drawings are not necessarily drawn to scale unless specifically indicated.

The word "exemplary" is used exclusively herein to mean "serving as an example, embodiment, or illustration. Any embodiment described herein as "exemplary" is not necessarily to be construed as preferred or advantageous over other embodiments.

The term "and/or" herein is merely an association describing an associated object, meaning that three relationships may exist, e.g., a and/or B, may mean: a exists alone, A and B exist simultaneously, and B exists alone. In addition, the term "at least one" herein means any one of a plurality or any combination of at least two of a plurality, for example, including at least one of A, B, C, and may mean including any one or more elements selected from the group consisting of A, B and C.

Furthermore, in the following detailed description, numerous specific details are set forth in order to provide a better understanding of the present disclosure. It will be understood by those skilled in the art that the present disclosure may be practiced without some of these specific details. In some instances, methods, means, elements and circuits that are well known to those skilled in the art have not been described in detail so as not to obscure the present disclosure.

Fig. 1 shows a flow chart of a neural network training method according to an embodiment of the present disclosure, as shown in fig. 1, the neural network training method includes:

in step S11, inputting the sample dynamic visual information of the sample scene into a first reconstruction network for processing, and obtaining a first reconstruction result, where the first reconstruction result includes a multi-level first feature map and a first reconstructed image;

in step S12, inputting a sample color image of the sample scene into a second reconstruction network for processing, and obtaining a second reconstruction result, where the second reconstruction result includes a multi-level second feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time;

determining a composite network loss of the first reconstruction network and the second reconstruction network according to the sample color image, the first reconstruction result and the second reconstruction result in step S13;

training the first reconstruction network and the second reconstruction network according to the integrated network loss in step S14,

wherein the first reconstruction network is configured to generate a color image from dynamic visual information, and the second reconstruction network is configured to train the first reconstruction network.

According to the neural network training method disclosed by the embodiment of the disclosure, the first reconstruction result of the sample dynamic visual information obtained by the first reconstruction network is close to the second reconstruction result of the color image obtained by the second reconstruction network through training, so that the first reconstruction network can obtain the color image with higher trueness degree based on the sample dynamic visual information. Because the frequency of the dynamic visual information is higher than the acquisition frequency of the color image, the acquisition frequency of the color image can be improved by processing the dynamic visual information through the first reconstruction network, the tracking of the motion track or the motion of a moving object is facilitated, and the tracking effect is improved.

In one possible implementation, the neural network training method may be performed by an electronic device such as a terminal device or a server, the terminal device may be a User Equipment (UE), a mobile device, a User terminal, a cellular phone, a cordless phone, a Personal Digital Assistant (PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, or the like, and the method may be implemented by a processor calling a computer readable instruction stored in a memory. Alternatively, the method may be performed by a server.

In one possible implementation, Dynamic visual receptors (DVS) are sensitive to the rate of change of light intensity, and each pixel may record the amount of change in light intensity at the pixel location, and when the amount of change exceeds a threshold, a positive or negative going pulse is generated, i.e., Dynamic visual information.

For example, an Event Camera (Event Camera) is a dynamic visual receptor that can be used to capture the rate of change of light intensity for a preset scene. When a target in a preset scene is abnormal or performs some actions, the light intensity of the target presented in the event camera can change to a certain degree, and the event camera can sharply capture the change to obtain dynamic visual information.

In one possible implementation, the frame rate of the dynamic visual receptors is higher than that of a normal camera or webcam, for example, the frame rate of a camera or a conventional webcam is about 100fps, while the frame rate of the dynamic visual receptors is about 1,000,000 fps. Therefore, in the time interval between two frames of images shot by a common camera or a camera, a plurality of frames of dynamic visual information can be shot.

In one possible implementation, since the acquisition frequency of the color image is low, it is difficult to capture the color image of the target object in the time interval between the acquisition of two frames of color images, and the target object is also tracked by the color image. While the frequency of acquiring the dynamic visual information is high, a plurality of dynamic visual information can be acquired in the time interval between the acquisition of two frames of color images, and therefore, a more color image can be acquired based on the dynamic visual information, for example, in the time interval between the shooting of two frames of color images, a color image is generated by the dynamic visual information for tracking the target object in the time interval.

In one possible implementation, although the dynamic visual information is obtained frequently, the amount of information in the dynamic visual information of a single frame is small, and the pixel data is sparse. The characteristic extraction of the dynamic visual information is difficult to obtain characteristic maps with rich information, and the reconstruction of color images based on the characteristic maps is also difficult.

In one possible implementation, based on the above problem, the dynamic visual information may be processed through a first reconstruction network for processing the dynamic visual information to generate a feature map from the dynamic visual information. The first reconstructed network may be trained prior to processing using the first reconstructed network. For example, the first reconstruction network may be aided in training by a second reconstruction network for processing color images. In an example, a first reconstruction result obtained by processing the dynamic visual information by the first reconstruction network may be approximated to a second reconstruction result obtained by processing the color image by the second reconstruction network, that is, a feature map obtained by processing the dynamic visual information by the first reconstruction network and the generated color image are approximated to a feature map and a color image obtained by processing the color image by the second reconstruction network, respectively, so that the first reconstruction network can generate an image close to a real color image. In an example, the first reconstruction network comprises a recurrent neural network and the second reconstruction network comprises a convolutional neural network. The present disclosure does not limit the type of the first reconstruction network and the second reconstruction network.

In one possible implementation, the training may be performed by sample dynamic visual information and sample color images of the same scene. The method can acquire a plurality of sample color images and sample dynamic visual information for the same sample scene at the same time, and train by using the sample color images and the sample dynamic visual information acquired at the same time.

In one possible implementation, in step S11, the first reconstruction network may include an encoding sub-network and a decoding sub-network, the encoding sub-network may include a plurality of network levels and may be configured to acquire feature information of the dynamic visual information, and the decoding sub-network may include a plurality of network levels and may be configured to perform decoding based on the feature information and acquire the first reconstruction result, that is, obtain the multi-level first feature map (i.e., the feature map output by each level of the decoding sub-network) and the first reconstruction image.

In a possible implementation manner, in step S12, the second reconstruction network may also include an encoding sub-network and a decoding sub-network, the encoding sub-network may include a plurality of network levels and may be configured to obtain feature information of the color image, and the decoding sub-network may include a plurality of network levels and may be configured to perform decoding based on the feature information and obtain the second reconstruction result, that is, obtain the multi-level second feature map (i.e., the feature map output by each level of the decoding sub-network) and the second reconstruction image.

In one possible implementation, the purpose of the training is to enable the first reconstruction network to generate an image consistent with the color image based on the dynamic visual information, and thus, the first reconstructed image generated by the first reconstruction network may be more realistic and accurate by reducing a difference between the first reconstruction result of the first reconstruction network and the second reconstruction result of the second reconstruction network and reducing a difference between the second reconstruction image and the color image. That is, the difference between each level of the first feature map in the first reconstruction result and the corresponding level of the second feature map in the second reconstruction result is reduced, and the difference between the first reconstructed image and the second reconstructed image is reduced, and the difference between the second reconstructed image and the color image is reduced. In this way, the accuracy of the feature map output by each level of the first reconstruction network can be improved, thereby further improving the fidelity and accuracy of the image generated by the first reconstruction network.

In one possible implementation, before training, both the first reconstruction network and the second reconstruction network may have errors, i.e., the first reconstruction result output by the first reconstruction network and the second reconstruction result output by the second reconstruction network are not consistent, and/or the second reconstruction image output by the second reconstruction network is not consistent with the sample color image.

In one possible implementation, in step S13, the integrated network loss of the first reconstruction network and the second reconstruction network may be determined based on the sample color image, the first reconstruction result, and the second reconstruction result. As described above, there may be a difference between the first reconstruction result and the second reconstruction result, and there may be a difference between the second reconstruction image and the sample color image, and the network loss may be determined based on the difference, and the network loss may be reduced step by step through training to reduce the difference, so as to achieve the training purpose.

In one possible implementation, step S13 may include: determining a first network loss according to the first reconstruction result and the second reconstruction result; determining a second network loss from the second reconstructed image and the sample color image; and determining the comprehensive network loss according to the first network loss and the second network loss.

In one possible implementation, the first network loss may be determined based on the first reconstruction result and the second reconstruction result, and reducing the first network loss during the training may reduce a difference between the first reconstruction result and the second reconstruction result. The first reconstruction result and the second reconstruction result can both comprise a multi-level feature map and a reconstructed image, and the difference between the feature maps of the corresponding levels and the difference between the reconstructed images can be respectively determined so as to determine the first network loss.

In one possible implementation, the first reconstruction result may include n-level feature maps, the second reconstruction result may include m-level feature maps, m and n are positive integers, and m and n may be equal or unequal. Feature maps with the same resolution (i.e., feature maps with the same scale) may be selected from the first reconstruction result and the second reconstruction result, and the sub-losses of each stage may be determined based on the feature maps with the same resolution, for example, if the resolution of the first feature map at level 1 in the first reconstruction result is equal to the resolution of the second feature map at level 2 in the second reconstruction result, and the resolution of the first feature map at level 2 in the first intermediate result is equal to the resolution of the second feature map at level 4 in the second reconstruction result, then the first sub-loss may be determined based on the first feature map at level 1 in the first reconstruction result and the second feature map at level 2 in the second reconstruction result, and the first sub-loss may be determined based on the first feature map at level 1 in the first reconstruction result and the second feature map at level 2 in the second reconstruction result.

In an example, the first reconstruction result includes a number of levels of the first feature map equal to a number of levels of the second feature map included in the second reconstruction result, and the first feature map and the second feature map of each stage have equal resolutions. For example, the number of network levels of the decoding subnetworks of the first reconstruction network is equal to the number of network levels of the decoding subnetworks of the second reconstruction network, and the upsampling rate of each network level is equal. In this case, the first sub-penalties of the first and second feature maps may be determined separately for each level. In an example, the first sub-penalty may be determined based on a difference between the first feature map and a second feature map having the same resolution, e.g., the first sub-penalty may be determined based on a difference between a pixel value of each pixel point in the first feature map and a pixel value of a corresponding pixel point in the second feature map. The present disclosure does not limit the manner in which the first sub-loss is determined.

In one possible implementation, the first and second reconstruction results further include reconstructed images, respectively, and the first and second reconstructed images may have equal resolutions and may have equal resolutions to the sample color image. The second sub-loss may be determined based on a difference between the first reconstructed image and the second reconstructed image, e.g., the second sub-loss may be determined based on a difference between a pixel value of each pixel point in the first reconstructed image and a pixel value of a corresponding pixel point in the second reconstructed image. The present disclosure does not limit the manner in which the second sub-loss is determined.

In a possible implementation manner, the first sub-losses and the second sub-losses at each stage may be subjected to weighted summation processing to obtain the first network loss. In the training process, the loss of the first network is reduced, so that each level of feature map output by the first reconstruction network after the dynamic visual information is processed is closer to the feature map output by the second reconstruction network after the color image is processed, that is, the feature map corresponding to the dynamic visual information is closer to the feature map corresponding to the color image, and the reconstructed image obtained based on the dynamic visual information is closer to the real color image.

In a possible implementation manner, the training may make the feature map corresponding to the dynamic visual information and the feature map corresponding to the color image closer, and may also make the second reconstructed image output by the second reconstruction network by processing the color image closer to the real color image. By the method, the accuracy and the fidelity of the second reconstructed image can be improved, and then the accuracy and the fidelity of the first reconstructed image close to the second reconstructed image can be improved, namely, the first reconstructed image can obtain an accurate and vivid reconstructed image.

In one possible implementation, the second network loss may be determined based on a difference between the second reconstructed image and the true sample color image, for example, the second network loss may be determined based on a difference between a pixel value of each pixel point in the second reconstructed image and a pixel value of a corresponding pixel point in the sample color image. The present disclosure does not limit the manner in which the second network loss is determined.

In a possible implementation manner, the first network loss and the second network loss may be integrated, so that the first feature map output by the first reconstruction network can gradually approach the second feature map of the real sample color image in the training process, and the first reconstruction image output by the first reconstruction network can gradually approach the real sample color image.

In one possible implementation, determining the composite network loss according to the first network loss and the second network loss includes: and carrying out weighted summation processing on the first network loss and the second network loss to obtain the comprehensive network loss. Namely, the first network loss and the second network loss are weighted and summed, and the total network loss obtained by the weighted and summed is trained so that the total network loss is gradually reduced, that is, each stage of the first feature map is made to approach the corresponding second feature map, and the first reconstructed image is made to approach the second reconstructed image, and the second reconstructed image is made to approach the sample color image. That is, by reducing the difference between the first feature map and the feature map of the true sample color image, the first reconstructed image is made more realistic and accurate, that is, the first reconstructed image is made closer to the sample color image.

In a possible implementation manner, the training steps may be iteratively performed, and when a training condition is satisfied, the trained first reconstruction network and the trained second reconstruction network are obtained. The training conditions may include a training frequency condition, that is, when the above training steps are performed iteratively for a preset number of times, the training is completed. Alternatively, the training condition may include whether the integrated network loss is less than or equal to a preset threshold or converges to a preset interval, and the training may be completed if the integrated network loss is less than or equal to the preset threshold or converges to the preset interval.

In one possible implementation, the dynamic visual information may be processed through the trained first reconstruction network to generate a realistic and accurate color image.

In one possible implementation, the present disclosure also relates to an image generation method, including: and inputting the dynamic visual information of the preset scene acquired at a plurality of moments in a first time period into a first reconstruction network trained according to the neural network training method for processing, and generating a first color image corresponding to each dynamic visual information.

In an example, the length of the first time period may be equal to a time interval between two frames of color images (e.g., images or video frames) of the preset scene captured by the camera or the camera, or may be a time interval between multiple frames of color images of the preset scene captured. That is, the start-stop time of the first period may be the time when the color image is captured.

In another example, the start-stop time of the first time period may not be the time when the color image is acquired, and the length of the first time period may also be less than the time period between the two frames of color images acquired by the camera or the camera, so that only one frame of color image needs to be acquired in the first time period. The length and the starting time of the first time period are not limited by the present disclosure. For example, the start time of the first period may be before one frame of color image is captured, and the end time of the first period may be after one frame of color image is captured, and does not necessarily coincide with the time at which the color image is captured.

In one possible implementation, the method further includes: and obtaining a video of the preset scene in the first time period according to the first color image and a second color image of the preset scene obtained in the first time period, wherein the obtaining frequency of the dynamic visual information is higher than that of the color image. That is, the first color image generated by the plurality of dynamic visual information acquired in the first time period may complement the second color image captured in the first time period, that is, the number of color images acquired in the first time period may be increased, so that the time interval between the color images in the first time period may be shorter, and a more accurate video may be acquired. The observation and tracking effects of the target object in the preset scene are improved, for example, the motion and/or motion track of the acquired target object can be more accurate.

In a possible implementation manner, the second color image may not be captured in the first time period, and the plurality of first color images are generated only by the captured dynamic visual information, and constitute a video of the preset scene in the first time period. The present disclosure does not limit whether the second color image is photographed.

According to the neural network training method disclosed by the embodiment of the disclosure, the characteristic diagram corresponding to the dynamic visual information is closer to the characteristic diagram corresponding to the color image through training, and the second reconstructed image output by processing the color image through the second reconstruction network is closer to the real color image, that is, the accuracy and the fidelity of the second reconstructed image can be improved, so that the accuracy and the fidelity of the first reconstructed image close to the second reconstructed image can be improved. Because the frequency of the dynamic visual information is higher than the acquisition frequency of the color image, the acquisition frequency of the color image can be improved by processing the dynamic visual information through the first reconstruction network, the tracking of the motion track or the motion of a moving object is facilitated, and the tracking effect is improved.

Fig. 2 is a schematic diagram illustrating an application of a neural network training method according to an embodiment of the present disclosure, and as shown in fig. 2, a first reconstruction network and a second reconstruction network may be trained by sample dynamic visual information and sample color images of the same scene.

In one possible implementation, the decoding sub-network of the first reconstruction network includes three levels, and can output two levels of the first feature map and the first reconstructed image. The decoding sub-network of the second reconstruction network comprises three levels and can output two levels of second feature maps and second reconstruction images. And the resolution of each level of the first characteristic map is equal to that of the second characteristic map, and the resolution of the first reconstructed image, the resolution of the second reconstructed image and the resolution of the sample color image are equal.

In one possible implementation, the first sub-loss of the first stage 1 may be determined from a difference between the first stage first feature map and the first stage second feature map, and the first sub-loss of the second stage 2 may be determined from a difference between the second stage first feature map and the second stage second feature map.

In one possible implementation, the second sub-loss 3 may be determined based on a difference between the first reconstructed image and the second reconstructed image. The first network loss can be obtained by weighted summation of the first sub-loss (loss 1), the first sub-loss (loss 2) and the second sub-loss (loss 3).

In one possible implementation, the second network loss, loss4, may be determined from the difference between the second reconstructed image and the sample color image. And carrying out weighted summation on the first network loss and the second network loss to obtain the comprehensive network loss of the first reconstruction network and the second reconstruction network. In an example, the integrated network loss can be determined by the following equation 1:

Loss＝w₁×Loss₁+w₂×Loss₂+w₃×Loss₃+w₄×Loss₄ (1)

wherein, w₁、w₂、w₃And w₄Is a preset weight value.

In one possible implementation, the first reconstruction network and the second reconstruction network may be trained by synthesizing network losses, and the training may be completed when the synthesized network losses are less than or equal to a preset threshold or converge within a preset interval. The trained first reconstruction network can be used for processing the dynamic visual information to obtain a color image.

In one possible implementation, the neural network training method may be used to obtain a first reconstruction network capable of processing dynamic visual information to generate color images, and more color images may be generated by the first reconstruction network, such that the number of color images obtained within a predetermined time period is increased and the time interval of the color images is smaller, facilitating observation and tracking of the target. The present disclosure does not limit the application field of the neural network training method.

Fig. 3 shows a block diagram of a neural network training device according to an embodiment of the present disclosure, as shown in fig. 3, including: the first reconstruction module 11 is configured to input sample dynamic visual information of a sample scene into a first reconstruction network for processing, so as to obtain a first reconstruction result, where the first reconstruction result includes a multi-level first feature map and a first reconstruction image; a second reconstruction module 12, configured to input a sample color image of the sample scene into a second reconstruction network for processing, so as to obtain a second reconstruction result, where the second reconstruction result includes a multi-level second feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time; a loss determining module 13, configured to determine, according to the sample color image, the first reconstruction result, and the second reconstruction result, a comprehensive network loss of the first reconstruction network and the second reconstruction network; a training module 14, configured to train the first reconstruction network and the second reconstruction network according to the comprehensive network loss, where the first reconstruction network is used to generate a color image according to dynamic visual information, and the second reconstruction network is used to train the first reconstruction network.

The present disclosure also relates to an image generation apparatus comprising: and the generating module is used for inputting the dynamic visual information of the preset scene acquired at a plurality of moments in a first time period into the first reconstruction network trained by the neural network training method for processing, and generating a first color image corresponding to each dynamic visual information.

It is understood that the above-mentioned method embodiments of the present disclosure can be combined with each other to form a combined embodiment without departing from the logic of the principle, which is limited by the space, and the detailed description of the present disclosure is omitted. Those skilled in the art will appreciate that in the above methods of the specific embodiments, the specific order of execution of the steps should be determined by their function and possibly their inherent logic.

In addition, the present disclosure also provides a neural network training device, an electronic device, a computer-readable storage medium, and a program, which can be used to implement any one of the neural network training methods provided by the present disclosure, and the corresponding technical solutions and descriptions and corresponding descriptions in the methods section are not repeated.

In some embodiments, functions of or modules included in the apparatus provided in the embodiments of the present disclosure may be used to execute the method described in the above method embodiments, and specific implementation thereof may refer to the description of the above method embodiments, and for brevity, will not be described again here.

Embodiments of the present disclosure also provide a computer-readable storage medium having stored thereon computer program instructions, which when executed by a processor, implement the above-mentioned method. The computer readable storage medium may be a non-volatile computer readable storage medium.

An embodiment of the present disclosure further provides an electronic device, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to invoke the memory-stored instructions to perform the above-described method.

The disclosed embodiments also provide a computer program product comprising computer readable code, which when run on a device, a processor in the device executes instructions for implementing the neural network training method provided in any of the above embodiments.

The embodiments of the present disclosure also provide another computer program product for storing computer readable instructions, which when executed, cause a computer to perform the operations of the neural network training method provided in any one of the embodiments.

The electronic device may be provided as a terminal, server, or other form of device.

Fig. 4 illustrates a block diagram of an electronic device 800 in accordance with an embodiment of the disclosure. For example, the electronic device 800 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, or the like terminal.

Referring to fig. 4, electronic device 800 may include one or more of the following components: processing component 802, memory 804, power component 806, multimedia component 808, audio component 810, input/output (I/O) interface 812, sensor component 814, and communication component 816.

The processing component 802 generally controls overall operation of the electronic device 800, such as operations associated with display, telephone calls, data communications, camera operations, and recording operations. The processing components 802 may include one or more processors 820 to execute instructions to perform all or a portion of the steps of the methods described above. Further, the processing component 802 can include one or more modules that facilitate interaction between the processing component 802 and other components. For example, the processing component 802 can include a multimedia module to facilitate interaction between the multimedia component 808 and the processing component 802.

The memory 804 is configured to store various types of data to support operations at the electronic device 800. Examples of such data include instructions for any application or method operating on the electronic device 800, contact data, phonebook data, messages, pictures, videos, and so forth. The memory 804 may be implemented by any type or combination of volatile or non-volatile memory devices such as Static Random Access Memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic or optical disks.

The power supply component 806 provides power to the various components of the electronic device 800. The power components 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power for the electronic device 800.

The multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and a user. In some embodiments, the screen may include a Liquid Crystal Display (LCD) and a Touch Panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive an input signal from a user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense an edge of a touch or slide action, but also detect a duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 808 includes a front facing camera and/or a rear facing camera. The front camera and/or the rear camera may receive external multimedia data when the electronic device 800 is in an operation mode, such as a shooting mode or a video mode. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.

The audio component 810 is configured to output and/or input audio signals. For example, the audio component 810 includes a Microphone (MIC) configured to receive external audio signals when the electronic device 800 is in an operational mode, such as a call mode, a recording mode, and a voice recognition mode. The received audio signals may further be stored in the memory 804 or transmitted via the communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

The I/O interface 812 provides an interface between the processing component 802 and peripheral interface modules, which may be keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to: a home button, a volume button, a start button, and a lock button.

The sensor assembly 814 includes one or more sensors for providing various aspects of state assessment for the electronic device 800. For example, the sensor assembly 814 may detect an open/closed state of the electronic device 800, the relative positioning of components, such as a display and keypad of the electronic device 800, the sensor assembly 814 may also detect a change in the position of the electronic device 800 or a component of the electronic device 800, the presence or absence of user contact with the electronic device 800, orientation or acceleration/deceleration of the electronic device 800, and a change in the temperature of the electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 814 may also include an acceleration sensor, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.

The communication component 816 is configured to facilitate wired or wireless communication between the electronic device 800 and other devices. The electronic device 800 may access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 816 receives a broadcast signal or broadcast related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 816 further includes a Near Field Communication (NFC) module to facilitate short-range communications. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID) technology, infrared data association (IrDA) technology, Ultra Wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

In an exemplary embodiment, the electronic device 800 may be implemented by one or more Application Specific Integrated Circuits (ASICs), Digital Signal Processors (DSPs), Digital Signal Processing Devices (DSPDs), Programmable Logic Devices (PLDs), Field Programmable Gate Arrays (FPGAs), controllers, micro-controllers, microprocessors or other electronic components for performing the above-described methods.

In an exemplary embodiment, a non-transitory computer-readable storage medium, such as the memory 804, is also provided that includes computer program instructions executable by the processor 820 of the electronic device 800 to perform the above-described methods.

Fig. 5 illustrates a block diagram of an electronic device 1900 in accordance with an embodiment of the disclosure. For example, the electronic device 1900 may be provided as a server. Referring to fig. 5, electronic device 1900 includes a processing component 1922 further including one or more processors and memory resources, represented by memory 1932, for storing instructions, e.g., applications, executable by processing component 1922. The application programs stored in memory 1932 may include one or more modules that each correspond to a set of instructions. Further, the processing component 1922 is configured to execute instructions to perform the above-described method.

The electronic device 1900 may further include a power component 1926 configured to perform power management of the electronic device 1900, a wired or wireless networkA network interface 1950 is configured to connect the electronic device 1900 to a network, and an input/output (I/O) interface 1958. The electronic device 1900 may operate based on an operating system, such as Windows Server, stored in memory 1932^TM，Mac OS X^TM，Unix^TM,Linux^TM，FreeBSD^TMOr the like.

In an exemplary embodiment, a non-transitory computer readable storage medium, such as the memory 1932, is also provided that includes computer program instructions executable by the processing component 1922 of the electronic device 1900 to perform the above-described methods.

The present disclosure may be systems, methods, and/or computer program products. The computer program product may include a computer-readable storage medium having computer-readable program instructions embodied thereon for causing a processor to implement various aspects of the present disclosure.

The computer readable storage medium may be a tangible device that can hold and store the instructions for use by the instruction execution device. The computer readable storage medium may be, for example, but not limited to, an electronic memory device, a magnetic memory device, an optical memory device, an electromagnetic memory device, a semiconductor memory device, or any suitable combination of the foregoing. More specific examples (a non-exhaustive list) of the computer readable storage medium would include the following: a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a Static Random Access Memory (SRAM), a portable compact disc read-only memory (CD-ROM), a Digital Versatile Disc (DVD), a memory stick, a floppy disk, a mechanical coding device, such as punch cards or in-groove projection structures having instructions stored thereon, and any suitable combination of the foregoing. Computer-readable storage media as used herein is not to be construed as transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission medium (e.g., optical pulses through a fiber optic cable), or electrical signals transmitted through electrical wires.

The computer-readable program instructions described herein may be downloaded from a computer-readable storage medium to a respective computing/processing device, or to an external computer or external storage device via a network, such as the internet, a local area network, a wide area network, and/or a wireless network. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers. The network adapter card or network interface in each computing/processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium in the respective computing/processing device.

The computer program instructions for carrying out operations of the present disclosure may be assembler instructions, Instruction Set Architecture (ISA) instructions, machine-related instructions, microcode, firmware instructions, state setting data, or source or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a Local Area Network (LAN) or a Wide Area Network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet service provider). In some embodiments, the electronic circuitry that can execute the computer-readable program instructions implements aspects of the present disclosure by utilizing the state information of the computer-readable program instructions to personalize the electronic circuitry, such as a programmable logic circuit, a Field Programmable Gate Array (FPGA), or a Programmable Logic Array (PLA).

Various aspects of the present disclosure are described herein with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the disclosure. It will be understood that each block of the flowchart illustrations and/or block diagrams, and combinations of blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer-readable program instructions.

These computer-readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions/acts specified in the flowchart and/or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer-readable storage medium that can direct a computer, programmable data processing apparatus, and/or other devices to function in a particular manner, such that the computer-readable medium storing the instructions comprises an article of manufacture including instructions which implement the function/act specified in the flowchart and/or block diagram block or blocks.

The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other devices to cause a series of operational steps to be performed on the computer, other programmable apparatus or other devices to produce a computer implemented process such that the instructions which execute on the computer, other programmable apparatus or other devices implement the functions/acts specified in the flowchart and/or block diagram block or blocks.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and/or flowchart illustration, and combinations of blocks in the block diagrams and/or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The computer program product may be embodied in hardware, software or a combination thereof. In an alternative embodiment, the computer program product is embodied in a computer storage medium, and in another alternative embodiment, the computer program product is embodied in a Software product, such as a Software Development Kit (SDK), or the like.

Having described embodiments of the present disclosure, the foregoing description is intended to be exemplary, not exhaustive, and not limited to the disclosed embodiments. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen in order to best explain the principles of the embodiments, the practical application, or improvements made to the technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

1. A neural network training method, comprising:

inputting sample dynamic visual information of a sample scene into a first reconstruction network for processing to obtain a first reconstruction result, wherein the first reconstruction result comprises a multi-stage first feature map and a first reconstruction image;

inputting a sample color image of the sample scene into a second reconstruction network for processing to obtain a second reconstruction result, wherein the second reconstruction result comprises a multi-level second feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time;

determining the comprehensive network loss of the first reconstruction network and the second reconstruction network according to the sample color image, the first reconstruction result and the second reconstruction result;

training the first reconstruction network and the second reconstruction network based on the synthetic network loss,

2. The method of claim 1, wherein determining a composite network loss for the first reconstruction network and the second reconstruction network from the sample color image, the first reconstruction result, and the second reconstruction result comprises:

determining a first network loss according to the first reconstruction result and the second reconstruction result;

determining a second network loss from the second reconstructed image and the sample color image;

and determining the comprehensive network loss according to the first network loss and the second network loss.

3. The method of claim 2, wherein determining a first network loss based on the first reconstruction result and the second reconstruction result comprises:

respectively determining first sub-losses of each stage according to the first characteristic diagram of each stage and a second characteristic diagram with the same resolution as the first characteristic diagram of each stage;

determining a second sub-loss according to the first reconstructed image and the second reconstructed image;

and carrying out weighted summation processing on the first sub-losses and the second sub-losses of each stage to obtain the first network loss.

4. The method of claim 1, wherein the first reconstruction network comprises a recurrent neural network and the second reconstruction network comprises a convolutional neural network.

5. An image generation method, comprising:

inputting dynamic visual information of a preset scene acquired at a plurality of moments in a first time period into a first reconstruction network for processing, and generating a first color image corresponding to each dynamic visual information, wherein the first reconstruction network is obtained by training according to the neural network training method of any one of claims 1 to 5.

6. The method of claim 5, further comprising:

and obtaining a video of the preset scene in the first time period according to the first color image and a second color image of the preset scene obtained in the first time period, wherein the obtaining frequency of the dynamic visual information is higher than that of the color image.

7. A neural network training device, comprising:

the first reconstruction module is used for inputting the sample dynamic visual information of the sample scene into a first reconstruction network for processing to obtain a first reconstruction result, and the first reconstruction result comprises a multi-stage first feature map and a first reconstruction image;

the second reconstruction module is used for inputting a sample color image of the sample scene into a second reconstruction network for processing to obtain a second reconstruction result, the second reconstruction result comprises a second multi-level feature map and a second reconstruction image, and the sample color image and the sample dynamic visual information are obtained at the same time;

a loss determining module, configured to determine, according to the sample color image, the first reconstruction result, and the second reconstruction result, a comprehensive network loss of the first reconstruction network and the second reconstruction network;

a training module for training the first reconstruction network and the second reconstruction network based on the synthetic network loss,

8. An image generation apparatus, comprising:

a generating module, configured to input dynamic visual information of a preset scene, acquired at multiple times within a first time period, into a first reconstruction network for processing, and generate a first color image corresponding to each piece of dynamic visual information, where the first reconstruction network is obtained by training according to the neural network training device of claim 7.

9. An electronic device, comprising:

a processor;

a memory for storing processor-executable instructions;

wherein the processor is configured to invoke the memory-stored instructions to perform the method of any of claims 1 to 6.

10. A computer readable storage medium having computer program instructions stored thereon, which when executed by a processor implement the method of any one of claims 1 to 6.