Point cloud understanding method and device, equipment, storage medium and program product

Through the point cloud understanding method, the point cloud encoder and text description model are used to obtain the object's text description information in extreme environments, solving the problem of information acquisition failure caused by too low light intensity and ensuring the normal acquisition of object information.

CN118982829AActive Publication Date: 2024-11-19GD MIDEA AIR CONDITIONING EQUIP CO LTD +1
View PDF 9 Cites 0 Cited by

Patent Information

Application Number
CN202411062793.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-08-02
Publication Date
2024-11-19
Estimated Expiration
2044-08-02

AI Technical Summary

Technical Problem

In extreme environments, such as the ambient light intensity is too low, the object information collected through the image may not be clear enough, resulting in the object information acquisition failure.

Method used

Using point cloud understanding method, by obtaining point cloud data collected by point cloud acquisition equipment, processing the data based on the point cloud encoder and text description model, and generating text description information of the object, including object type, orientation, shape and size, etc. information.

Benefits of technology

In extreme environments, point cloud data can be collected normally, ensuring the acquisition of object information, and providing clear text description information, thus solving the problem of failure in image acquisition.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118982829A_ABST
    Figure CN118982829A_ABST
Patent Text Reader

Abstract

The embodiment of the invention provides a point cloud understanding method and device, equipment and a storage medium, and belongs to the technical field of artificial intelligence vision. According to the method, collected point cloud data in a certain scene is processed, and text description information of an object (any person or object) in the scene is obtained. Therefore, in some extreme environments in which clear images cannot be acquired, such as environments with too low ambient light intensity, the point cloud data can be normally acquired generally, and then the text description information of the object in the scene can still be acquired through the method, so that normal acquisition of object information is ensured.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence visual technology, and in particular to a method, device, equipment, storage medium and program product for point cloud understanding. Background Art

[0002] With the development of artificial intelligence vision technology, computers can perform object detection based on images and obtain meaningful object information, where the object can be any person or object. For example, the computer extracts text description information of an object (such as a vehicle on the side) from the image captured by the car camera, and the driver determines the driving operation based on the text description information. For another example, a drone collects images of a place that is not suitable for people to enter (such as a gas-contaminated area), and the computer extracts text description information of an object (such as a device suspected of leaking gas) in the image to determine the situation of the place.

[0003] However, the images collected by the above method in some extreme environments (eg, the ambient light intensity is too low) may not be clear enough, resulting in failure in obtaining object information. Summary of the invention

[0004] In order to solve the related technical problems, the present disclosure provides a method, device, equipment, storage medium and program product for point cloud understanding. The technical solution is as follows:

[0005] In a first aspect, a method for understanding a point cloud is provided, the method comprising:

[0006] Obtain point cloud data collected by point cloud collection equipment;

[0007] Processing the point cloud data based on a point cloud encoder to obtain feature data corresponding to the point cloud data;

[0008] Processing the feature data based on a text description model to obtain text description information of at least one target object corresponding to the point cloud data, wherein the text description information at least includes an object type;

[0009] Output text description information of the at least one target object.

[0010] In a possible implementation, the method further includes:

[0011] Acquire a training sample, wherein the training sample includes sample point cloud data and corresponding reference text description information;

[0012] Based on the training samples, the point cloud encoder to be trained and the text description model to be trained are trained.

[0013] In a possible implementation, obtaining a training sample includes:

[0014] Get the benchmark text description information;

[0015] Based on the point cloud generation model, generating point cloud data corresponding to the reference text description information as sample point cloud data;

[0016] A training sample is determined based on the sample point cloud data and the reference text description information.

[0017] In a possible implementation, the generating point cloud data corresponding to the reference text description information based on the point cloud generation model as sample point cloud data includes:

[0018] Adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information;

[0019] Inputting the plurality of expanded reference text description information into the point cloud generation model respectively to obtain a plurality of point cloud data output by the point cloud generation model as a plurality of sample point cloud data;

[0020] The determining of training samples based on the sample point cloud data and the reference text description information includes:

[0021] A plurality of sample point cloud data are respectively combined with the reference text description information to obtain a plurality of training samples.

[0022] In a possible implementation, adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information includes:

[0023] The reference text description information is input into a large language model, so that different extended description information is added to the reference text description information through the large language model, thereby obtaining a plurality of extended reference text description information output by the large language model.

[0024] In a possible implementation, adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information includes:

[0025] Different extended description information is randomly added to the reference text description information to obtain a plurality of extended reference text description information.

[0026] In a possible implementation manner, the text description information further includes at least one of object orientation, object shape, and object size.

[0027] In a second aspect, a device for understanding point cloud is provided, the device comprising:

[0028] An acquisition module, used to acquire point cloud data collected by a point cloud acquisition device;

[0029] A processing module, configured to process the point cloud data based on a point cloud encoder to obtain feature data corresponding to the point cloud data; and to process the feature data based on a text description model to obtain text description information of at least one target object corresponding to the point cloud data, wherein the text description information at least includes an object type;

[0030] The output module is used to output the text description information of the at least one target object.

[0031] In a possible implementation, the device further includes a training module, which is used to:

[0032] Acquire a training sample, wherein the training sample includes sample point cloud data and corresponding reference text description information;

[0033] Based on the training samples, the point cloud encoder to be trained and the text description model to be trained are trained.

[0034] In a possible implementation, the training module is used to:

[0035] Get the benchmark text description information;

[0036] Based on the point cloud generation model, generating point cloud data corresponding to the reference text description information as sample point cloud data;

[0037] A training sample is determined based on the sample point cloud data and the reference text description information.

[0038] In a possible implementation, the training module is used to:

[0039] Adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information;

[0040] Inputting the plurality of expanded reference text description information into the point cloud generation model respectively to obtain a plurality of point cloud data output by the point cloud generation model as a plurality of sample point cloud data;

[0041] The determining of training samples based on the sample point cloud data and the reference text description information includes:

[0042] A plurality of sample point cloud data are respectively combined with the reference text description information to obtain a plurality of training samples.

[0043] In a possible implementation, the training module is used to:

[0044] The reference text description information is input into a large language model, so that different extended description information is added to the reference text description information through the large language model, thereby obtaining a plurality of extended reference text description information output by the large language model.

[0045] In a possible implementation, the training module is used to:

[0046] Different extended description information is randomly added to the reference text description information to obtain a plurality of extended reference text description information.

[0047] In a possible implementation manner, the text description information further includes at least one of object orientation, object shape, and object size.

[0048] In a third aspect, a computer device is provided, comprising a memory and a processor, wherein the memory is used to store computer instructions, and the processor executes the computer instructions stored in the memory so that the computer device executes the method provided by the first aspect and its possible implementation methods.

[0049] In a fourth aspect, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program code. When the computer program code is executed by a computer device, the computer device executes the method provided by the first aspect and its possible implementation methods.

[0050] In a fifth aspect, a computer program product is provided. The computer program product includes a computer program code. When the computer program code is executed by a computer device, the computer device executes the method provided by the first aspect and its possible implementation methods.

[0051] In the present disclosure, the collected point cloud data of a certain scene is processed to obtain text description information of the object (any person or object) in the scene. In this way, in some extreme environments where clear images cannot be obtained, such as environments with low ambient light intensity, point cloud data can generally be collected normally, and then the text description information of the object in the scene can still be obtained through the above method, thereby ensuring the normal acquisition of object information. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1 is a structural diagram of a server provided by an embodiment of the present disclosure;

[0053] Figure 2 It is a schematic diagram of a processing flow of a method for understanding point cloud provided by an embodiment of the present disclosure;

[0054] Figure 3 is a schematic diagram of a training processing flow provided by an embodiment of the present disclosure;

[0055] Figure 4 is a schematic diagram of a processing flow for obtaining training samples provided by an embodiment of the present disclosure;

[0056] Figure 5 It is a structural schematic diagram of a point cloud understanding device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0057] In order to make the objectives, technical solutions and advantages of the present disclosure more clear, the embodiments of the present disclosure will be further described in detail below with reference to the accompanying drawings.

[0058] The disclosed embodiment provides a method for understanding point clouds. The executor of the method may be a server. The server may be a single server or a server group. If it is a single server, the server may be responsible for all processing in the following scheme. If it is a server group, different servers in the server group may be responsible for different processing in the following scheme. The specific processing distribution may be arbitrarily set by technical personnel according to actual needs and will not be elaborated here.

[0059] Figure 1 is a schematic diagram of the structure of a server provided by an embodiment of the present disclosure. From the perspective of hardware composition, the structure of the server can be as follows Figure 1 As shown, it includes a processor 110 , a memory 120 , a communication component 130 and a display component 140 .

[0060] The processor 110 may be a central processing unit (CPU) or a system on chip (SoC), etc. The processor 110 may be used to process various operation instructions, for example, processing point cloud data based on a point cloud encoder, etc.

[0061] The memory 120 may include various volatile memories or non-volatile memories, such as a solid state disk (SSD), a dynamic random access memory (DRAM), etc. The memory 120 may be used to store initial data, intermediate data, and result data used in the relevant processing process, for example, to store point cloud data.

[0062] The communication component 130 may be a wired network connector, ultra wide band (UWB), wireless fidelity (WiFi) module, Bluetooth module, cellular network communication module, etc. The communication component 130 may be used to transmit data with other devices, and the other devices may be other servers or terminals, etc. For example, to obtain point cloud data collected by a point cloud collection device.

[0063] The display component 140 may be an independent screen, or a screen integrated with the terminal body, a projector, etc. The screen may be a touch screen or a non-touch screen (which may be displayed via remote live broadcast). The display component 140 is used to display processing results, for example, display text description information, etc.

[0064] With the development of artificial intelligence vision technology, computers can detect objects based on images and obtain meaningful object information. For scenes that are inconvenient for people to enter or areas that are inconvenient for people to observe, cameras can be used to collect images instead of technicians.

[0065] For example, when a driver is driving a car, he needs to focus on the front of the car, which makes it easy to ignore other directions of the car. At this time, the car can be installed with multiple cameras to obtain images in all directions of the car. The car computer processes the images to obtain text description information, and the driver can drive according to the text description information corresponding to the images in all directions.

[0066] For example, when toxic gas pollution occurs in a certain area, it is necessary to check the equipment that leaks the toxic gas. Manual inspection is difficult, but a drone can be controlled to enter the toxic gas pollution area and take pictures of the equipment in the area. The captured images can be processed to obtain text description information, thereby identifying the equipment that leaked the toxic gas.

[0067] However, in some extreme environments (e.g., the ambient light intensity is too low), the captured images may not be clear enough, resulting in failure or inaccuracy in obtaining object information. In order to better obtain accurate object information, the present disclosure provides a method for understanding point clouds. The processing flow of the method can be as follows: Figure 2 As shown, the following steps are included:

[0068] 201, obtaining point cloud data collected by a point cloud collection device.

[0069] The point cloud acquisition device may be a laser scanner, which uses the principle of laser ranging to record the three-dimensional coordinates of a large number of points on the object being measured.

[0070] 202 , processing the point cloud data based on the point cloud encoder to obtain feature data corresponding to the point cloud data.

[0071] The point cloud data is encoded by a point cloud encoder to obtain feature data corresponding to the point cloud data.

[0072] 203 : Process the feature data based on the text description model to obtain text description information of at least one target object corresponding to the point cloud data.

[0073] The feature data is input into the text description model to obtain text description information output by the text description model, wherein the text description information at least includes the object type.

[0074] The text description information also includes at least one of the following information: object type, object orientation, object shape, and object size. The point cloud encoder and the text description model can form a point cloud-text understanding (point contrastive captioner, PointCoca) model.

[0075] 204. Output text description information of at least one target object.

[0076] Output the text description information corresponding to the point cloud data, and display the point cloud data and text description information at the same time.

[0077] In addition, the PointCoca model can also realize the functions of point cloud data search and point cloud data classification. When the user wants to search for point cloud data, he can enter the text description information to obtain the point cloud data output by the PointCoca model. For example, if the input text description information is "a chubby teddy bear with a oval-shaped body and oversized ears", all point cloud data that meet the text description information can be output. When the user wants to classify point cloud data, he can enter multiple point cloud data and obtain the classified point cloud data output by the PointCoca model. For example, if the input multiple point cloud data are two teddy bears and three tables, the PointCoca model encodes the multiple point cloud data respectively and classifies them according to the distance. The output result is point cloud data with category labels.

[0078] In one possible implementation, the training process can be as follows: Figure 3 As shown, the following steps are included:

[0079] 301, obtain training samples.

[0080] The training samples include sample point cloud data and corresponding benchmark text description information.

[0081] There are many ways to obtain training samples. For example, you can directly obtain an existing sample set including point cloud data and text description information, or you can input text description information into the point cloud generation model to obtain the point cloud data output by the point cloud generation model. In this way, the text description information and the corresponding point cloud data are used as a training sample. For example, the text description information "a teddy bear" and the point cloud data of a teddy bear can be used as a training sample.

[0082] 302 , based on the training samples, the point cloud encoder to be trained and the text description model to be trained are trained.

[0083] The sample point cloud data is input into the point cloud encoder to be trained to obtain the predicted feature data corresponding to the sample point cloud data output by the point cloud encoder to be trained. The point cloud data can be downsampled by sampling the farthest point, and then multiple point clusters are obtained by using the k-nearest neighbor algorithm. The multiple point clusters are sorted based on the center point position, and then feature extraction is performed on each point cluster to obtain point cloud features, and then the point cloud features are encoded to obtain predicted feature data.

[0084] The predicted feature data is processed based on the text description model to be trained to obtain predicted text description information of at least one target object corresponding to the sample point cloud data. A loss value is determined based on the predicted text description information and the sample text description information corresponding to the sample point cloud data. Parameters of the point cloud encoder to be trained and the text description model to be trained are adjusted based on the loss value. If the training end condition is met, a trained point cloud encoder and a trained text description model are obtained.

[0085] In a possible implementation, the process of obtaining training samples can be as follows: Figure 4 As shown, the following steps are included:

[0086] 401, obtain the reference text description information.

[0087] The technician inputs the reference text description information, and the reference text description information at least includes the object type.

[0088] 402 , based on the point cloud generation model, generate point cloud data corresponding to the reference text description information as sample point cloud data.

[0089] The reference text description information is input into the point cloud generation model to obtain point cloud data output by the point cloud generation model as sample point cloud data. Alternatively, multiple extended reference text description information may be generated based on the reference text description information, and the extended reference text description information is then input into the point cloud generation model to obtain multiple point cloud data output by the point cloud generation model as sample point cloud data.

[0090] Different extended description information may be added to the base text description information to obtain a plurality of extended base text description information. The following are two methods for generating the extended base text description information.

[0091] Method 1: Input the base text description information into the large language model, and the large language model adds different extended description information to the base text description information to obtain multiple extended base text description information output by the large language model. For example, if the base text description information is "a teddy bear", the output of the large language model may be "a teddy bear holding a flower", "a chubby teddy bear with a oval-shaped body and oversized ears", etc.

[0092] Method 2: randomly adding different extended description information to the baseline text description information to obtain multiple extended baseline text description information. When selecting training samples, text description information with the same information (such as object type, object shape, etc.) can be selected as the baseline text description information. When randomly adding different extended description information, at least one information (information other than the baseline text description information, such as object size, object orientation, etc.) can be selected as the extended description information to obtain multiple extended baseline text description information. For example, the object type included in the baseline text description information is "teddy bear", and the information included in the extended description information is the object size. In this way, multiple extended baseline text description information including object type and object size are obtained, wherein the object type is all "teddy bear" and the object sizes are different.

[0093] The multiple expanded reference text description information is respectively input into the point cloud generation model to obtain multiple point cloud data output by the point cloud generation model as multiple sample point cloud data. For example, the multiple expanded reference text description information includes object type and object size, wherein the object type is all "teddy bear" and the object size is different. The multiple point cloud data output by the point cloud generation model are also all "teddy bear" and the object size is different.

[0094] 403 , determining a training sample based on the sample point cloud data and the reference text description information.

[0095] Combine multiple sample point cloud data with the reference text description information to obtain multiple training samples. For example, the reference text description information "a teddy bear" and the sample point cloud data of a teddy bear can be used as a training sample, and the reference text description information "a teddy bear" and the sample point cloud data of another teddy bear can be used as a training sample.

[0096] In the disclosed embodiment, the collected point cloud data of a certain scene is processed to obtain text description information of the object (any person or object) in the scene. In this way, in some extreme environments where it is impossible to obtain a clear image, such as an environment with too low ambient light intensity, the point cloud data can generally be collected normally, and then the text description information of the object in the scene can still be obtained through the above method, thereby ensuring the normal acquisition of object information.

[0097] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0098] Based on the same technical concept, the embodiment of the present disclosure also provides a device for understanding point clouds, such as Figure 5 As shown, the device comprises:

[0099] The acquisition module 510 is used to acquire the point cloud data collected by the point cloud acquisition device, and can specifically implement the processing function of the above step 201 and other implicit steps.

[0100] The processing module 520 is used to process the point cloud data based on the point cloud encoder to obtain feature data corresponding to the point cloud data; and to process the feature data based on the text description model to obtain text description information of at least one target object corresponding to the point cloud data, wherein the text description information at least includes the object type. Specifically, the processing functions of the above steps 202 and 203 and other implicit steps can be realized.

[0101] The output module 530 is used to output text description information of at least one target object, and can specifically implement the processing function of the above step 204 and other implicit steps.

[0102] In a possible implementation, the device further includes a training module 540, which is used to: obtain training samples, wherein the training samples include sample point cloud data and corresponding reference text description information; based on the training samples, train the point cloud encoder to be trained and the text description model to be trained. Specifically, the processing functions of the above steps 301 and 302, as well as other implicit steps, can be realized.

[0103] In a possible implementation, the training module 540 is used to: obtain the reference text description information; generate point cloud data corresponding to the reference text description information as sample point cloud data based on the point cloud generation model; and determine the training sample based on the sample point cloud data and the reference text description information. Specifically, the processing functions of the above steps 401, 402, and 403, as well as other implicit steps, can be implemented.

[0104] In a possible implementation, the training module 540 is used to: add different extended description information to the reference text description information to obtain multiple extended reference text description information; input the multiple extended reference text description information into the point cloud generation model to obtain multiple point cloud data output by the point cloud generation model as multiple sample point cloud data; combine the multiple sample point cloud data with the reference text description information to obtain multiple training samples. Specifically, the processing function of the above step 402 and other implicit steps can be realized.

[0105] In a possible implementation, the training module 540 is used to: input the reference text description information into the large language model, so as to add different extended description information to the reference text description information through the large language model, and obtain a plurality of extended reference text description information output by the large language model. Specifically, the processing function of the above step 402 and other implicit steps can be implemented.

[0106] In a possible implementation, the training module 540 is used to randomly add different extended description information to the reference text description information to obtain a plurality of extended reference text description information. Specifically, the processing function of the above step 402 and other implicit steps can be implemented.

[0107] In a possible implementation manner, the text description information further includes at least one of object orientation, object shape, and object size.

[0108] The acquisition module 510, processing module 520, output module 530, and training module 540 may be implemented by a processor, or by a processor in conjunction with a memory and a display.

[0109] In the disclosed embodiment, the collected point cloud data of a certain scene is processed to obtain text description information of the object (any person or object) in the scene. In this way, in some extreme environments where it is impossible to obtain a clear image, such as an environment with too low ambient light intensity, the point cloud data can generally be collected normally, and then the text description information of the object in the scene can still be obtained through the above method, thereby ensuring the normal acquisition of object information.

[0110] When the point cloud understanding device provided in the above embodiment is in operation, only the division of the above functional modules is used as an example. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the electric device is divided into different functional modules to complete all or part of the functions described above. In addition, the point cloud understanding device provided in the above embodiment and the point cloud understanding method embodiment belong to the same concept. The specific implementation process is detailed in the method embodiment and will not be repeated here.

[0111] The present disclosure also provides a computer-readable storage medium. The computer-readable storage medium may be any available medium that can be stored by a computer device or a data storage device such as a data center that includes one or more available media. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk). The computer-readable storage medium includes instructions that instruct a computing device to perform a method for understanding a point cloud.

[0112] 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 it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the protection scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for point cloud understanding, It is characterized in that The method comprises: Obtain point cloud data collected by point cloud collection equipment; Processing the point cloud data based on a point cloud encoder to obtain feature data corresponding to the point cloud data; Processing the feature data based on a text description model to obtain text description information of at least one target object corresponding to the point cloud data, wherein the text description information at least includes an object type; Output text description information of the at least one target object.

2. The method according to claim 1, It is characterized in that The method further comprises: Acquire a training sample, wherein the training sample includes sample point cloud data and corresponding reference text description information; Based on the training samples, the point cloud encoder to be trained and the text description model to be trained are trained.

3. The method according to claim 2, It is characterized in that The obtaining of training samples comprises: Get the benchmark text description information; Based on the point cloud generation model, generating point cloud data corresponding to the reference text description information as sample point cloud data; A training sample is determined based on the sample point cloud data and the reference text description information.

4. The method according to claim 3, It is characterized in that The step of generating point cloud data corresponding to the reference text description information based on the point cloud generation model as sample point cloud data includes: Adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information; Inputting the plurality of expanded reference text description information into the point cloud generation model respectively, to obtain a plurality of point cloud data output by the point cloud generation model as a plurality of sample point cloud data; The determining of training samples based on the sample point cloud data and the reference text description information includes: A plurality of sample point cloud data are respectively combined with the reference text description information to obtain a plurality of training samples.

5. The method according to claim 4, It is characterized in that The adding different extended description information to the reference text description information to obtain a plurality of extended reference text description information includes: The reference text description information is input into a large language model, so that different extended description information is added to the reference text description information through the large language model, thereby obtaining a plurality of extended reference text description information output by the large language model.

6. The method according to claim 1, It is characterized in that The text description information also includes at least one of object orientation, object shape, and object size.

7. A device for understanding point clouds, It is characterized in that The device comprises: An acquisition module, used to acquire point cloud data collected by a point cloud acquisition device; A processing module, configured to process the point cloud data based on a point cloud encoder to obtain feature data corresponding to the point cloud data; and to process the feature data based on a text description model to obtain text description information of at least one target object corresponding to the point cloud data, wherein the text description information at least includes an object type; The output module is used to output the text description information of the at least one target object.

8. A computer device, It is characterized in that The computer device comprises a memory and a processor, wherein the memory is used to store computer instructions; The processor executes the computer instructions stored in the memory to enable the computer device to perform the method according to any one of claims 1 to 6.

9. A computer-readable storage medium, It is characterized in that The computer-readable storage medium stores a computer program code. When the computer program code is executed by a computer device, the computer device executes the method according to any one of claims 1 to 6.

10. A computer program product, It is characterized in that The computer program product comprises a computer program code. When the computer program code is executed by a computer device, the computer device executes the method according to any one of claims 1 to 6.

Citation Information

Patent Citations

  • Laser-point cloud based method for extracting mountainous road by use of three-dimensional characteristic description model

    CN106022259A

  • Point cloud data processing method and device, computer storage medium and electronic equipment

    CN112115744A

  • Single-stage three-dimensional scene point cloud target positioning and description method

    CN116664983A

  • Image generation method and device, electronic equipment and storage medium

    CN116797684A

  • Three-dimensional pre-training large model parameter efficient fine tuning method and system

    CN117252986A