CN-120726543-B - Video data processing method, device and system for self-adaptive evolution

CN120726543BCN 120726543 BCN120726543 BCN 120726543BCN-120726543-B

Abstract

The application discloses a video data processing method, device and system with self-adaptive evolution, and relates to the field of electric digital data processing. The method comprises the steps of determining a data distribution feature vector of target video data, retrieving a target model parameter set in an action space according to the data distribution feature vector and a preset model framework, wherein the action space comprises a plurality of groups of selectable model parameter sets, generating model parameter description information according to the preset model framework and the target model parameter set, and sending the model parameter description information to terminal equipment, wherein the model parameter description information is used for adjusting a video data processing model deployed in the terminal equipment to obtain the target video data processing model. The application solves the technical problem of low model data processing efficiency caused by adopting a static data processing model in the related technology.

Inventors

SUN HAO
WU DONG
LI XUELONG
HE ZHONGJIANG
ZHOU XUYANG

Assignees

中电信人工智能科技(北京)有限公司

Dates

Publication Date: 20260512
Application Date: 20250828

Claims (11)

1. A method of adaptively evolving video data processing, comprising: determining a data distribution feature vector of target video data, wherein the data distribution feature vector is used for representing a first data feature of the target video data, and the first data feature comprises at least one of category distribution entropy, a data processing difficulty evaluation result and spatial resolution, and the data processing difficulty evaluation result is determined by the following steps: Determining a second data characteristic of the target video data, wherein the second data characteristic comprises at least one of image entropy, gradient magnitude; Processing the second data features through a difficulty classifier to obtain the data processing difficulty evaluation result of the target video data, wherein the data processing difficulty evaluation result comprises difficult processing data and easy processing data; Searching in an action space according to the data distribution feature vector and a preset model framework to obtain a target model parameter set, wherein the action space comprises a plurality of groups of selectable model parameter sets; Generating model parameter description information according to the preset model architecture and the target model parameter set; And sending the model parameter description information to terminal equipment, wherein the model parameter description information is used for adjusting a video data processing model deployed in the terminal equipment to obtain a target video data processing model.
2. The method for adaptively evolving video data processing according to claim 1, further comprising: Determining a metadata processing model, and training the metadata processing model by adopting a data feature vector in a preset knowledge base; After training the metadata processing model, sending a first model parameter set of the metadata processing model to each terminal device, wherein the terminal device adjusts parameters of the target video data processing model deployed locally according to the first model parameter set, and trains the adjusted target video data processing model by using local data after adjustment to obtain a second model parameter set, and the second model parameter set comprises model parameters of the trained target video data processing model; Receiving a second model parameter set sent by each terminal device and data processing difficulty information of local data of the terminal device; Determining weights corresponding to the second model parameter sets of each group according to the data processing difficulty information, and carrying out weighted aggregation processing on the second model parameter sets of each group according to the weights to obtain a third model parameter set; And sending the third model parameter set to each terminal device, wherein the third model parameter set is used for updating the target video data processing model deployed on the local terminal device.
3. The method according to claim 2, wherein the predetermined knowledge base is configured to determine a similarity between the data feature vectors stored in the predetermined knowledge base, and to combine the data feature vectors having the similarity greater than a predetermined similarity threshold, the method further comprising: receiving a video data sample acquired by the terminal equipment, and determining a data feature vector of the video data sample; And storing the data characteristic vector of the video data sample into the preset knowledge base.
4. The method for adaptively evolving video data processing according to claim 1, further comprising: receiving a video data sample acquired by the terminal equipment, and storing the video data sample into a dynamic negative sample pool; And after the number of the video data samples stored in the dynamic negative sample pool meets a preset requirement, adjusting the preset model framework according to the video data samples in the dynamic negative sample pool.
5. The method for adaptively evolving video data processing according to claim 1, further comprising: determining the data processing difficulty evaluation result of each frame of image in the target video data; Determining computing power resource allocation information corresponding to the image according to the data processing difficulty evaluation result, wherein the computing power resource allocation information comprises computing power resources allocated when the terminal equipment calls the target video data processing model to process the image; and calling the target video processing model to process the target video data according to the computing power resource allocation information.
6. The video data processing method of claim 1, wherein the target video data processing model includes a three-dimensional convolution block therein, wherein the three-dimensional convolution block is configured to capture cross-frame motion features of the target video data from a spatial dimension and a temporal dimension.
7. An adaptively evolving video data processing device, comprising: the first processing module is used for determining a data distribution feature vector of target video data, wherein the data distribution feature vector is used for representing first data features of the target video data, and the first data features comprise at least one of category distribution entropy, a data processing difficulty evaluation result and spatial resolution, and the data processing difficulty evaluation result is determined by the following modes: Determining a second data characteristic of the target video data, wherein the second data characteristic comprises at least one of image entropy, gradient magnitude; Processing the second data features through a difficulty classifier to obtain the data processing difficulty evaluation result of the target video data, wherein the data processing difficulty evaluation result comprises difficult processing data and easy processing data; The second processing module is used for retrieving a target model parameter set in an action space according to the data distribution feature vector and a preset model framework, wherein the action space comprises a plurality of groups of selectable model parameter sets; The third processing module is used for generating model parameter description information according to the preset model architecture and the target model parameter set; and the fourth processing module is used for sending the model parameter description information to the terminal equipment, wherein the model parameter description information is used for adjusting a video data processing model deployed in the terminal equipment to obtain a target video data processing model.
8. An adaptive evolving video data processing system, comprising a server and a plurality of terminal devices, wherein, The server is used for determining a data distribution feature vector of target video data, wherein the data distribution feature vector is used for representing first data features of the target video data, the first data features comprise at least one of category distribution entropy, a data processing difficulty evaluation result and spatial resolution, the data processing difficulty evaluation result is determined by determining second data features of the target video data, wherein the second data features comprise at least one of image entropy and gradient amplitude, processing the second data features through a difficulty classifier to obtain the data processing difficulty evaluation result of the target video data, wherein the data processing difficulty evaluation result comprises difficult data and easy-to-process data, retrieving in an action space according to the data distribution feature vector and a preset model architecture to obtain a target model parameter set, wherein the action space comprises a plurality of groups of selectable model parameter sets, generating model parameter description information according to the preset model parameter set and the target model parameter set, and sending the model parameter description information to the terminal equipment; The terminal equipment is used for adjusting a video data processing model deployed in the terminal equipment according to the model parameter description information to obtain a target video data processing model, and the target video data processing model is used for processing the target video data.
9. A non-volatile storage medium, wherein a program is stored in the non-volatile storage medium, wherein the program, when run, controls a device in which the non-volatile storage medium is located to perform the adaptively evolving video data processing method according to any one of claims 1 to 6.
10. An electronic device comprising a memory and a processor for executing a program stored in the memory, wherein the program is executed to perform the adaptively evolving video data processing method of any one of claims 1 to 6.
11. A computer program product comprising a computer program which, when executed by a processor, implements the adaptively evolving video data processing method according to any one of claims 1 to 6.

Description

Video data processing method, device and system for self-adaptive evolution Technical Field The application relates to the field of electric digital data processing, in particular to a video data processing method, device and system with self-adaptive evolution. Background In the related art, when processing video data, it is common to determine a static video data processing model, and then use the model to process the video data. The problem with this approach is that when there is a discrepancy between the data to be processed and the training data set, it can result in inefficient data processing. In view of the above problems, no effective solution has been proposed at present. Disclosure of Invention The embodiment of the application provides a video data processing method, device and system with self-adaptive evolution, which at least solve the technical problem of low model data processing efficiency caused by adopting a static data processing model in the related technology. According to one aspect of the embodiment of the application, a video data processing method for self-adaptive evolution is provided, which comprises the steps of determining a data distribution feature vector of target video data, retrieving a target model parameter set in an action space according to the data distribution feature vector and a preset model framework, wherein the action space comprises a plurality of groups of selectable model parameter sets, generating model parameter description information according to the preset model framework and the target model parameter set, and sending the model parameter description information to terminal equipment, wherein the model parameter description information is used for adjusting a video data processing model deployed in the terminal equipment to obtain the target video data processing model. Optionally, the data distribution feature vector is used for representing a first data feature of the target video data, the first data feature comprises at least one of category distribution entropy, a data processing difficulty evaluation result and spatial resolution, wherein the data processing difficulty evaluation result is determined by determining a second data feature of the target video data, the second data feature comprises at least one of image entropy and gradient amplitude, and the second data feature is processed through a difficulty classifier to obtain the data processing difficulty evaluation result of the target video data, and the data processing difficulty evaluation result comprises difficult processing data and easy processing data. The method comprises the steps of determining a metadata processing model, training the metadata processing model by adopting a data feature vector in a preset knowledge base, after training the metadata processing model, sending a first model parameter set of the metadata processing model to each terminal device, wherein the terminal device adjusts parameters of a target video data processing model deployed locally according to the first model parameter set, trains the adjusted target video data processing model by utilizing local data after adjustment to obtain a second model parameter set, the second model parameter set comprises model parameters of the target video data processing model after training, receiving the second model parameter set sent by each terminal device and data processing difficulty information of local data of the terminal device, determining weights corresponding to each group of second model parameter sets according to the data processing difficulty information, carrying out weighted aggregation processing on each group of second model parameter sets according to the weights to obtain a third model parameter set, and sending the third model parameter set to each terminal device, wherein the third model parameter set is used for updating the target video data processing model deployed locally at the terminal device. Optionally, the preset knowledge base is used for determining similarity between data feature vectors stored in the preset knowledge base and combining the data feature vectors with similarity greater than a preset similarity threshold, and the method further comprises the steps of receiving video data samples collected by the terminal equipment, determining the data feature vectors of the video data samples, and storing the data feature vectors of the video data samples in the preset knowledge base. Optionally, the method further comprises the steps of receiving video data samples collected by the terminal equipment, storing the video data samples in the dynamic negative-sample pool, and adjusting a preset model framework according to the video data samples in the dynamic negative-sample pool after the number of the video data samples stored in the dynamic negative-sample pool meets preset requirements. Optionally, the method further comprises the steps of determining a data processing difficulty