Multi-angle image acquisition system for LoRA model training

Abstract

The invention discloses a multi-angle image acquisition system for LoRA model training, which relates to the technical field of image acquisition and comprises an acquisition module, an evaluation module, a preprocessing module and a feature extraction module. According to the invention, the definition and illumination uniformity of the image collected in real time are calculated and evaluated through the evaluation module, the illumination uniformity and the image definition of the image are compared with the set threshold value through the Python script, and when the illumination uniformity and the image definition of the image do not meet the preset condition, the image can be detected. The information is transmitted to the cloud server, the camera is started through the control module to shoot an object image at the angle again, and when the illumination uniformity and the image definition of the image meet preset conditions, the information is transmitted to the preprocessing module, so that it is ensured that data entering a subsequent LoRA model training link has stability and reliability in quality; the LoRA model is helped to learn features more accurately, and the training effect and precision are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of image acquisition, and particularly relates to a multi-angle image acquisition system for LoRA model training. Background Art

[0002] With the rapid development of information technology, data acquisition for AI model training is commonly adopted in the current industrial field. With its powerful data analysis and prediction capabilities, the AI model has become the core driving force to promote industrial production towards automation and intelligence. As the cornerstone of AI model training, the quality and diversity of data directly determine the performance and generalization ability of the model. In order to enable the AI model to comprehensively and accurately learn and understand these complex process and physical characteristics, multi-angle image acquisition becomes particularly important.

[0003] In the prior art, when acquiring images of an object, single-angle image acquisition is usually adopted, resulting in only obtaining single-view information of the object and being unable to comprehensively display the characteristics of the object, which is far from enough for LoRA model training that needs to learn all-round characteristics of the object; at the same time, during the acquisition process, the image clarity or light uniformity in certain angles or scenarios is prone to be affected by focusing or jitter, resulting in the lack or inaccuracy of object feature information in this view, making it difficult for the subsequent model to accurately learn the features, affecting the training effect and accuracy, and causing the model to perform poorly in tasks such as recognition and generation.

[0004] Therefore, a multi-angle image acquisition system for LoRA model training is proposed to solve the above problems. Summary of the Invention

[0005] The main object of the present invention is to provide a multi-angle image acquisition system for LoRA model training to solve the problems raised in the above background.

[0006] To achieve the above object, the technical solution adopted by the present invention is: a multi-angle image acquisition system for LoRA model training, the system includes an acquisition module, an evaluation module, a preprocessing module, and a feature extraction module; The acquisition module is used to integrate multiple camera modules, a light source module, a rotation module, and a control module on a portable bracket, take multi-angle images of an object through the portable bracket, and transmit them to the cloud server in real time through a wireless transmission module; The evaluation module is used to evaluate the clarity and light uniformity of the multi-angle captured images through the cloud server, and compare the light uniformity and image clarity of the images with a set threshold using a Python script; The preprocessing module is used to clean, unify the format, and annotate the images that pass the clarity and light uniformity of the multi-angle captured images; The feature extraction module is used to process the preprocessed image through a convolution operation formula and extract features from the image according to the requirements of the LoRA model.

[0007] Further, the acquisition module includes a camera module, a light source module, a rotation module, and a control module; The camera module uses a camera to take pictures of items; The light source module uses a fill light to supplement light during photography; The rotation module includes a rotation unit and a recording unit; The rotation unit uses a rotating base for 360-degree rotation of the camera; The recording unit uses an angle sensor for angle adjustment and recording of the camera.

[0008] Further, the control module includes a control unit, a Wi-Fi unit, a storage unit, and a power supply unit; The control unit uses a microcontroller to control the rotation of the rotating base, the shooting of the camera, and the power of the fill light; The Wi-Fi unit uses a Wi-Fi module for data transmission and reception; The storage unit uses a MicrosSD card slot for data storage; The power supply unit uses a storage battery.

[0009] Further, the evaluation module includes a receiving unit, an evaluation unit, and a processing unit; The adjustment unit uses a cloud server to receive multi-angle images taken by the camera in real time.

[0010] Further, the evaluation unit is used to calculate the clarity and illumination uniformity of the multi-angle images taken by the camera in real time. The calculation formula for the clarity of the multi-angle images taken by the camera in real time is: ; where G represents the average gradient value, and the larger the average gradient value, the higher the clarity of the picture. and respectively represent the horizontal and vertical gradients of the image at , and M and N are the number of rows and columns of the image; Set the threshold of the image clarity to X ≤ 10. When the calculated is greater than X, it means that the image clarity meets the requirements. When the calculated is less than or equal to X, it means that the image clarity does not meet the requirements.

[0011] Further, the calculation formula for the illumination uniformity of the multi-angle images captured by the real-time receiving camera is: = ; where represents the variance, and too large a variance indicates too low illumination uniformity of the picture, represents the brightness value of the pixel at the position of the -th row and the -th column of the image, represents the average brightness of all pixels in the image, and M and N are the number of rows and columns of the image; The threshold of the illumination uniformity is set to greater than or equal to 0.1. When the calculated is greater than or equal to Q, it means that the illumination uniformity of the image does not meet the requirements. When the calculated is less than or equal to Q, it means that the illumination uniformity of the image meets the requirements.

[0012] Further, the processing unit uses a Python script to compare the illumination uniformity and image clarity of the image with the set thresholds. When the illumination uniformity and image clarity of the image do not meet the preset conditions, the information is transmitted to the cloud server and the camera is started through the control module to re-capture the image of the item at this angle. When the illumination uniformity and image clarity of the image meet the preset conditions, the information is transmitted to the preprocessing module.

[0013] Further, the preprocessing module includes a cleaning unit, a formatting unit, and an annotation unit.

[0014] Further, the cleaning unit uses an image processor to clean and enhance the image; The formatting unit uses an ARM architecture chip to unify the format of the cleaned image; The annotation unit uses an AI acceleration chip to perform annotation processing on the image with unified format.

[0015] Further, the feature extraction module is used to extract features from the preprocessed image using the convolution operation formula. The operation formula is: ; where, represents the feature value at i, j of the input feature map, X represents the input feature map, represents the pixel value or feature value at the coordinate in the input feature map, W represents the convolution kernel, represents the weight value of the -th row and the -th column of the convolution kernel.

[0016] The present invention has the following beneficial effects: 1. In the present invention, after the Wi-Fi module transmits the multi-angle pictures collected in real time to the cloud server, the evaluation module calculates and evaluates the clarity and illumination uniformity of the pictures collected in real time, and compares the illumination uniformity and image clarity of the images with the set thresholds through Python scripts. When the illumination uniformity and image clarity of the images do not meet the preset conditions, the information is transmitted to the cloud server and the camera is activated through the control module to retake the image of the item at this angle. When the illumination uniformity and image clarity of the images meet the preset conditions, the information is transmitted to the preprocessing module, ensuring the stability and reliability of the data quality entering the subsequent LoRA model training session, which helps the LoRA model learn features more accurately and improves the training effect and accuracy.

[0017] 2. In the present invention, multiple camera modules, light source modules, rotation modules, and control modules are integrated into the portable bracket. Multiple cameras are used for multi-angle shooting, combined with two groups of fill lights with fixed color temperatures to optimize and standardize the lighting effect, reduce shadow or overexposure phenomena. During the rotation process of the rotation module, it cooperates with the angle sensor and microcontroller to achieve automatic rotation shooting within 360 degrees or other preset angle ranges without manual intervention, improving efficiency and accuracy. And through the internal Wi-Fi module, the pictures collected in real time are transmitted to the cloud server. Through the integrated design and automatic rotation shooting mechanism, it can quickly and efficiently collect multi-angle image data of items, greatly reducing the time and cost of manual operation.

[0018] 3. In the present invention, by cleaning, enhancing, unifying the format, and annotating the pictures that pass the evaluation, the workload of manually cleaning, annotating, and adjusting the images is reduced, improving work efficiency. At the same time, through convolutional operations, key features such as edges, textures, and shapes in the images are automatically extracted, filtering out irrelevant information, enabling the subsequent LoRA model to focus on important features for learning, and improving the feature expression ability and generalization ability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flowchart of a multi-angle image acquisition system for LoRA model training according to the present invention; Figure 2 is a flowchart of the acquisition module of a multi-angle image acquisition system for LoRA model training according to the present invention; Figure 3 is a flowchart of the rotation module of a multi-angle image acquisition system for LoRA model training according to the present invention; Figure 4 is a flowchart of the control module of a multi-angle image acquisition system for LoRA model training according to the present invention; Figure 5 This is the flowchart of the preprocessing module of a multi-angle image acquisition system for LoRA model training according to the present invention; Figure 6 This is the flowchart of the evaluation module of a multi-angle image acquisition system for LoRA model training according to the present invention. Specific embodiments

[0020] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1 Please refer to Figures 1 - 6 , the present invention provides a technical solution: a multi-angle image acquisition system for LoRA model training, the system includes an acquisition module, an evaluation module, a preprocessing module and a feature extraction module; The acquisition module is used to integrate multiple camera modules, light source modules, rotation modules and control modules on a portable bracket, take multi-angle image shots of items through the portable bracket, and transmit them to the cloud server in real time through a wireless transmission module; The evaluation module is used to evaluate the clarity and light uniformity of the multi-angle captured images through the cloud server, and use Python scripts to compare the light uniformity and image clarity of the images with the set thresholds; The preprocessing module is used to clean, unify the format and label the images that pass the clarity and light uniformity of the multi-angle captured images; The feature extraction module is used to process the preprocessed images through the convolution operation formula, and extract features from the images according to the requirements of the LoRA model.

[0022] The acquisition module includes a camera module, a light source module, a rotation module and a control module; The camera module uses a camera to take pictures of items; The light source module uses a fill light to supplement light during shooting; The rotation module includes a rotation unit and a recording unit; The rotation unit uses a rotating base for 360-degree rotation of the camera; The recording unit uses an angle sensor for angle adjustment and recording of the camera.

[0023] The control module includes a control unit, a Wi-Fi unit, a storage unit and a power supply unit; The control unit uses a microcontroller to control the rotation of the rotating base, the shooting of the camera, and the power of the fill light; The Wi-Fi unit uses a Wi-Fi module for data transmission and reception; The storage unit uses a MicrosSD card slot for data storage; The power supply unit uses a storage battery.

[0024] Multiple camera modules, light source modules, rotation modules, and control modules are integrated into the portable bracket. The camera module uses a standard camera with more than 2 million pixels, supporting image acquisition with a resolution of 1920*1080px. The light source module uses two groups of fill lights with fixed color temperatures to optimize and standardize the lighting effect, reducing shadows or overexposure. For sampling objects with high-gloss reflective surfaces, the main light source uses a polarized light source to reduce reflection. The speed of the rotation module can be adjusted as needed to adapt to different shooting requirements, smoothly completing the shooting process, and cooperating with the angle sensor and the microcontroller in the control module to achieve automatic rotation shooting within 360 degrees or other preset angle ranges, ensuring the uniformity and traceability of the shooting angles, being able to automatically record the precise information of each angle without manual intervention, improving efficiency and accuracy, and transmitting the captured pictures to the cloud server in real time through the internal Wi-Fi module. Through the integrated design and automated rotation shooting mechanism, this system can quickly and efficiently collect multi-angle image data of items, greatly reducing the time and cost of manual operations.

[0025] Embodiment 2 Please refer to Figures 1 - 6 , the present invention provides a technical solution: a multi-angle image acquisition system for LoRA model training, the system includes an acquisition module, an evaluation module, a preprocessing module, and a feature extraction module; The acquisition module is used to integrate multiple camera modules, light source modules, rotation modules, and control modules into the portable bracket, perform multi-angle image shooting and acquisition of items through the portable bracket, and transmit them to the cloud server in real time through the wireless transmission module; The evaluation module is used to evaluate the clarity and lighting uniformity of the multi-angle captured images through the cloud server, and use Python scripts to compare the lighting uniformity and image clarity of the images with the set thresholds; The preprocessing module is used to clean, unify the format, and annotate the images that pass the clarity and lighting uniformity of the multi-angle captured images; The feature extraction module is used to process the preprocessed images through the convolution operation formula and extract features from the images according to the requirements of the LoRA model.

[0026] The evaluation module includes a receiving unit, an evaluation unit, and a processing unit; The adjustment unit uses a cloud server to receive multi-angle images captured by a camera in real time.

[0027] The evaluation unit is used to calculate the clarity and illumination uniformity of the multi-angle images captured by the camera in real time. The calculation formula for the clarity of the multi-angle images captured by the camera in real time is: ; where G represents the average gradient value, and the larger the average gradient value, the higher the clarity of the picture. and respectively represent the gradients of the image in the horizontal and vertical directions at the position, and M and N are the number of rows and columns of the image; Set the threshold of the image clarity to X ≤ 10. When the calculated is greater than X, it means that the image clarity meets the requirements. When the calculated is less than or equal to X, it means that the image clarity does not meet the requirements.

[0028] The calculation formula for the illumination uniformity of the multi-angle images captured by the camera in real time is: = ; where represents the variance, and too large variance means too low illumination uniformity of the picture. represents the brightness value of the pixel at the th row and the th column position of the image. represents the average brightness value of all pixels of the image, and M and N are the number of rows and columns of the image; Set the threshold of the illumination uniformity to ≥ 0.1. When the calculated is greater than or equal to Q, it means that the illumination uniformity of the image does not meet the requirements. When the calculated is less than or equal to Q, it means that the illumination uniformity of the image meets the requirements.

[0029] The processing unit is used to compare the illumination uniformity and image clarity of the image with the set thresholds using a Python script. When the illumination uniformity and image clarity of the image do not meet the preset conditions, the information is transmitted to the cloud server and the camera is started through the control module to re-capture the image of the item at this angle. When the illumination uniformity and image clarity of the image meet the preset conditions, the information is transmitted to the preprocessing module.

[0030] After the Wi-Fi module transmits the real-time captured images to the cloud server, the evaluation module calculates and evaluates the clarity and illumination uniformity of the real-time captured images, and compares the illumination uniformity and image clarity of the images with the set thresholds through a Python script. When the illumination uniformity and image clarity of the images do not meet the preset conditions, the information is transmitted to the cloud server and the camera is activated through the control module to re-capture the images of the items at this angle. When the illumination uniformity and image clarity of the images meet the preset conditions, the information is transmitted to the preprocessing module, ensuring the stability and reliability of the data quality entering the subsequent LoRA model training session, which helps the LoRA model learn features more accurately and improve the training effect and accuracy.

[0031] Implementation Three Please refer to Figures 1 - 6 , a multi-angle image acquisition system for LoRA model training, the system includes an acquisition module, an evaluation module, a preprocessing module and a feature extraction module; The acquisition module is used to integrate multiple camera modules, a light source module, a rotation module and a control module on a portable bracket, take multi-angle images of items through the portable bracket, and transmit them to the cloud server in real time through a wireless transmission module; The evaluation module is used to evaluate the clarity and illumination uniformity of the multi-angle captured images through the cloud server, and compare the illumination uniformity and image clarity of the images with the set thresholds using a Python script; The preprocessing module is used to clean, unify the format and annotate the images that pass the clarity and illumination uniformity of the multi-angle captured images; The feature extraction module is used to process the preprocessed images through a convolution operation formula and extract features from the images according to the requirements of the LoRA model.

[0032] The preprocessing module includes a cleaning unit, a format unit and an annotation unit.

[0033] The cleaning unit uses an image processor to clean and enhance the images; The format unit uses an ARM architecture chip to unify the format of the cleaned images; The annotation unit uses an AI acceleration chip to annotate the images with unified formats.

[0034] The feature extraction module is used to extract features from the preprocessed images through a convolution operation formula, and the operation formula is: ; Among them, represents the feature value of the input feature map at i,j, X represents the input feature map, Represents the pixel value or feature value at the coordinate in the input feature map, W represents the convolutional kernel, and represents the weight value of the row and

[0035] column of the convolutional kernel.

[0036] By preprocessing the pictures that pass the evaluation, using an image processor and a specific filtering algorithm to process the images, smooth the images and remove these noise points, make the images cleaner and clearer, and redistribute the gray values of the images, so that the gray histogram distribution of the images is more uniform, thereby enhancing the overall contrast of the images. Use the ARM architecture chip to unify the format, and through the AI acceleration chip, label the pictures with unified formats, reducing the workload of manual cleaning, labeling and adjustment of the images, improving the work efficiency. At the same time, automatically extract key features such as edges, textures, and shapes in the images through convolutional operations, filter out irrelevant information, let the LoRA model focus on important features for learning, and improve the feature expression ability and the generalization ability of the model.In the present invention, a multi-angle image acquisition system for LoRA model training integrates multiple camera modules, a light source module, a rotation module, and a control module on a portable bracket. The camera module uses a standard camera with more than 2 million pixels and supports image acquisition with a resolution of 1920*1080px. Multiple cameras are used for multi-angle shooting. The light source module uses two groups of fill lights with a fixed color temperature to optimize and standardize the lighting effect, reducing shadows or overexposure phenomena. For sampling objects with surface specular reflection materials, the main light source uses a polarized light source to reduce reflection. The speed of the rotation module can be adjusted according to needs to adapt to different shooting requirements and smoothly complete the shooting process. During the rotation process, it cooperates with an angle sensor and a microcontroller in the control module to achieve automatic rotation shooting within 360 degrees or other preset angle ranges without manual intervention, improving efficiency and accuracy. It quickly and efficiently acquires multi-angle image data of items and transmits the collected pictures to the cloud server in real time through the internal Wi-Fi module. The evaluation module calculates and evaluates the clarity and lighting uniformity of the real-time collected pictures, and a Python script compares the lighting uniformity and image clarity of the images with the set thresholds. When the lighting uniformity and image clarity of the images do not meet the preset conditions, the information is transmitted to the cloud server and the camera is activated through the control module to reshoot the item image at this angle. When the lighting uniformity and image clarity of the images meet the preset conditions, the information is transmitted to the preprocessing module, ensuring the stability and reliability of the data quality entering the subsequent LoRA model training link, helping the LoRA model to learn features more accurately, improving the training effect and accuracy. Then, the qualified pictures are preprocessed. Using an image processor, through a specific filtering algorithm, the images are processed to smooth the images and remove these noise points, making the images cleaner and clearer, and redistributing the gray values of the images so that the gray histogram distribution of the images is more uniform, thereby enhancing the overall contrast of the images. The ARM architecture chip is used to unify the format, and through the AI acceleration chip, the pictures with unified format are labeled, reducing the workload of manually cleaning, labeling, and adjusting the images, improving work efficiency. At the same time, key features such as edges, textures, and shapes in the images are automatically extracted through convolution operations, filtering out irrelevant information, allowing the LoRA model to focus on important features for learning, and improving the feature expression ability and the generalization ability of the model.

[0037] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article or device comprising a series of elements not only includes those elements, but also includes other elements not expressly listed, or further includes elements inherent to such process, method, article or device.

[0038] Although the embodiments of the present invention have been shown and described, those of ordinary skill in the art can understand that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-angle image acquisition system for LoRA model training, characterized in that: The system includes an acquisition module, an evaluation module, a preprocessing module and a feature extraction module; The acquisition module is used to integrate multiple camera modules, light source modules, rotation modules and control modules in a portable bracket, capture and acquire multi-angle images of objects through the portable bracket, and transmit them to the cloud server in real time through the wireless transmission module; The evaluation module is used to evaluate the clarity and illumination uniformity of images captured from multiple angles through a cloud server, and use a Python script to compare the illumination uniformity and image clarity of the image with a set threshold; The pre-processing module is used to clean, unify the format and annotate the images collected by multi-angle shooting through the image clarity and illumination uniformity; The feature extraction module is used to process the preprocessed image through the convolution operation formula and extract features of the image according to the requirements of the LoRA model.

2. A multi-angle image acquisition system for LoRA model training according to claim 1, characterized in that: The acquisition module includes a camera module, a light source module, a rotation module and a control module; The camera module uses a camera to take a picture of the object; The light source module uses a fill light to provide fill light when taking photos; The rotation module includes a rotation unit and a recording unit; The rotating unit adopts a rotating base for rotating the camera 360 degrees; The recording unit uses an angle sensor to adjust and record the angle of the camera.

3. A multi-angle image acquisition system for LoRA model training according to claim 2, characterized in that: The control module includes a control unit, a Wi-Fi unit, a storage unit and a power supply unit; The control unit adopts a microcontroller, which is used to control the rotation of the rotating base, the shooting of the camera and the power of the fill light; The Wi-Fi unit uses a Wi-Fi module for data transmission and reception; The storage unit uses a MicrosSD card slot for data storage; The power supply unit adopts a battery.

4. A multi-angle image acquisition system for LoRA model training according to claim 3, characterized in that: The evaluation module includes a receiving unit, an evaluation unit and a processing unit; The adjustment unit uses a cloud server to receive multi-angle images taken by a camera in real time.

5. A multi-angle image acquisition system for LoRA model training according to claim 4, characterized in that: The evaluation unit is used to calculate the clarity and illumination uniformity of the multi-angle images captured by the real-time receiving camera. The calculation formula for the clarity of the multi-angle images captured by the real-time receiving camera is: ； Where G represents the average gradient value, and the larger the average gradient value, the higher the image clarity. and Respectively represent the images in The horizontal and vertical gradients at , M and N are the number of rows and columns of the image; The threshold of image clarity is set to X less than or equal to 10. When it is greater than X, it means that the image clarity meets the requirements. When it is less than or equal to X, it means that the image clarity does not meet the requirements.

6. A multi-angle image acquisition system for LoRA model training according to claim 5, characterized in that: The calculation formula for the illumination uniformity of the multi-angle images captured by the real-time receiving camera is: = ； in represents variance, and a large variance means that the uniformity of image illumination is too low. The representative image is Row, No. The brightness value of the pixel at column position, Represents the average brightness of all pixels in the image, M and N are the number of rows and columns of the image; The threshold of illumination uniformity is set as Greater than or equal to 0.1, when the calculated When it is greater than or equal to Q, it means that the image illumination uniformity does not meet the requirements. When it is less than or equal to Q, it means that the image illumination uniformity meets the requirements.

7. A multi-angle image acquisition system for LoRA model training according to claim 6, characterized in that: The processing unit is used for the Python script to compare the illumination uniformity and image clarity of the image with the set threshold value. When the illumination uniformity and image clarity of the image do not meet the preset conditions, the information is transmitted to the cloud server and the camera is started through the control module to re-shoot the image of the object at that angle. When the illumination uniformity and image clarity of the image meet the preset conditions, the information is transmitted to the preprocessing module.

8. A multi-angle image acquisition system for LoRA model training according to claim 1, characterized in that: The preprocessing module includes a cleaning unit, a format unit and a labeling unit.

9. A multi-angle image acquisition system for LoRA model training according to claim 8, characterized in that: The cleaning unit uses an image processor to clean and enhance the image; The format unit uses an ARM architecture chip to unify the format of the cleaned image; The labeling unit uses an AI acceleration chip to label the images in a unified format.

10. A multi-angle image acquisition system for LoRA model training according to claim 1, characterized in that: The feature extraction module is used to extract features from the preprocessed image using a convolution operation formula, and the operation formula is: ； in, represents the eigenvalue of the input feature map at i, j, X represents the input feature map, Represents the coordinates in the input feature map as The pixel value or feature value at , W represents the convolution kernel, Represents the convolution kernel Row, No. The weight value of the column.

Images