Method and system for measuring bidirectional reflectance distribution function of digital asset material surface

Abstract

The present invention provides a method and system for measuring the bidirectional reflectance distribution function of a digital asset material surface. In the measurement process, a polarized light source is first used to emit polarized light to the surface of the material to be measured, and a polarized camera is used to collect the reflection of the light on the surface of the material to be measured, so as to obtain multiple material photos, and the multiple material photos have repeated areas. Then the repeated areas are extracted to obtain the bidirectional reflectance distribution function of each position, that is, multiple optional values, which are used as inputs, and the optimal estimated value of BRDF is output through a pre-trained neural network prediction model, and the similarity between the material to be measured and the historical material is combined to determine whether to perform a secondary test, and the neural network prediction model is fine-tuned and trained again with the optimal estimated value or the correction value result of the secondary test, and the total loss function of all processed materials is combined for comprehensive learning, so that the model can be applied to different materials, and high-precision measurement results can be obtained for anisotropic materials.

Description

Technical Field

[0001] The present invention relates to the technical field of reflection distribution function measurement, and in particular to a method and system for measuring bidirectional reflection distribution function of a digital asset material surface. Background Art

[0002] The bidirectional reflectance distribution function BRDF (Bidirectional Reflectance Distribution Function) is designed to describe the relationship between incident light and reflected light on a surface. For incident light in one direction, the surface will reflect light in all directions of the hemisphere on the surface. The proportion of reflection in different directions is different. We use the bidirectional reflectance distribution function BRDF to represent the proportional relationship between reflected light and incident light in a specified direction. BRDF is defined as the ratio of the reflectivity of the target in a certain direction to the illuminance in the incident direction.

[0003] The full name of PBR is Physically-Based Rendering. PBR material refers to material based on physical rendering. The formation of PBR material, especially under PBR, is to affect BRDF by giving properties such as roughness and metalness to finally obtain a material with color expression close to the physical world. The construction process of such material is complicated. For example, the high-precision native scanning PBR digital asset library PBRMAX announced to the world by the applicant continuously scans high-precision 3D digital assets from the real world and provides high-quality materials for 3D content creators from all walks of life through unified and strict production standards, including PBR texture maps that are close to the real physical properties of objects, and support diffuse reflection, normal, roughness, metalness, displacement, ambient occlusion, highlight, glossiness, concave seam and other maps to build a commercial digital asset material database.

[0004] The surface reflection distribution function of digital asset materials is a very important parameter in PBR materials, which directly affects the accuracy of PBR material production. The existing method of measuring the surface reflection distribution function of digital asset materials generally only involves photographing, scanning and measuring a single material, especially relying on imported expensive equipment. Each measurement is independent of each other, and each time a new material is measured, adjustments need to be made, and even the measurement equipment needs to be redesigned, which is very expensive. In this way, the measurement methods for the surface reflection distribution function of digital asset materials are designed for single measurements, and different materials need to design different measurement methods, which lack universality and versatility.

[0005] Prior art literature:

[0006] Patent document 1: CN114609094A is a method for measuring the inner surface BRDF based on the brightness of the endoscope image and the normal distribution of the inner surface of the hole. Summary of the invention

[0007] The purpose of the present invention is to provide a method and system for measuring the bidirectional reflectance distribution function of the surface of a digital asset material, which can be applied to different materials and can obtain high-precision measurement results for anisotropic materials.

[0008] Another object of the present invention is to provide a method for measuring the bidirectional reflectance distribution function of the surface of a digital asset material, which is capable of correcting the measurement results and continuously self-learning and improving, using historical experience between different measurements for cumulative learning, and continuously adjusting the model to improve the accuracy of the measurement results.

[0009] According to a first aspect of the present invention, a method for measuring the bidirectional reflectance distribution function of a digital asset material surface is provided, comprising the following steps:

[0010] Step 1: Use a polarized light source to emit polarized light to the surface of the material to be tested;

[0011] Step 2: Use a polarization camera to capture images of the irradiated surface of the material to be tested, collect the reflection of the light on the surface of the material to be tested, and obtain multiple material photos, wherein the multiple material photos have repeated areas;

[0012] Step 3, extracting repeated areas in multiple material photos;

[0013] Step 4: Based on multiple material photos, estimate the bidirectional reflectance distribution function of each position in the repeated area to obtain p*q bidirectional reflectance distribution functions, where p and q represent the number of position points in the repeated area and the number of material photos, respectively.

[0014] Step 5: Taking the q bidirectional reflectance distribution functions at each position in the repeated area as input, outputting the optimal estimated values ​​of the q bidirectional reflectance distribution functions through the pre-trained neural network prediction model;

[0015] Step 6, obtaining the similarity between the material to be tested and the historical material processed by the neural network prediction model: if the similarity is greater than or equal to the preset threshold, output the optimal estimate obtained in the aforementioned step 5 as the bidirectional reflectance distribution function measurement result of the material to be tested; if the similarity is lower than the preset threshold, proceed to step 7 for a secondary detection process;

[0016] Step 7: Calculate the variance between the optimal estimates of the q bidirectional reflectance distribution functions obtained. If the variance is less than or equal to the preset threshold σ sd , then the optimal estimate obtained in the above step 5 is output as the bidirectional reflectance distribution function measurement result of the material to be tested; if the variance is greater than the preset threshold σ sd, then control the second imaging acquisition of the irradiated surface of the material to be tested and estimate the output through the neural network prediction model, recalculate and judge whether the variance between the optimal estimated values ​​of the bidirectional reflectance distribution function is less than or equal to the preset threshold σ sd , if the variance is less than or equal to the preset threshold σ sd , then the optimal estimated value of the re-estimated output is output as the bidirectional reflectance distribution function measurement result of the material to be measured, otherwise, go to step 8 to calibrate the infrared wave measurement equipment;

[0017] Step 8. Use infrared wave measuring equipment to measure the surface of the material to be tested, calculate the bidirectional reflection distribution function of the material to be tested, and use this as a correction value. Use the p*q bidirectional reflection distribution functions obtained for the first or second time as optional values ​​to fine-tune the neural network prediction model. During the fine-tuning training process, comprehensive learning is performed by combining the total loss functions of all processed materials.

[0018] As an optional implementation, the training process of the pre-trained neural network prediction model includes:

[0019] Initialize a neural network with multiple convolutional layers and fully connected layers;

[0020] The neural network is trained using the collected and calibrated standard material data in the sample database, wherein the standard material data includes optional values ​​of bidirectional reflectance distribution functions of various material objects and calibrated standard values;

[0021] The neural network prediction model is obtained through multiple trainings, wherein the loss function of the training process is:

[0022]

[0023] In the formula, e n is the total estimated loss error of the nth material, b is the corrected standard value, is the bidirectional reflectance distribution function measurement value output by the neural network estimation, and m is the position point to be measured in a single material that needs to be estimated.

[0024] As an optional implementation, the total loss function is:

[0025]

[0026] Wherein, N represents N different materials.

[0027] According to a second aspect of the purpose of the present invention, a computer-readable medium storing software is also proposed, wherein the software includes instructions that can be executed by one or more computers, and the instructions, through such execution, enable the one or more computers to perform operations, and the operations include the process of the aforementioned method.

[0028] According to a third aspect of the present invention, a computer system is also provided, comprising:

[0029] one or more processors;

[0030] A memory stores operable instructions, wherein when the instructions are executed by the one or more processors, the one or more processors are caused to perform operations, wherein the operations include the flow of the aforementioned method.

[0031] Based on the various aspects of the objectives and exemplary technical solutions proposed above, the present invention aims to propose a high-precision self-improving material surface reflection distribution function measurement method based on lifelong machine learning, which can be applied to the measurement of material surface reflection distribution function of different materials. The material surface reflection distribution function estimation network based on lifelong machine learning can autonomously estimate the optimal material surface reflection distribution function measurement value based on multiple overlapping area images input, and the material surface reflection distribution function estimation network can self-evaluate and tune the training according to the learning results, continuously optimize and improve, and improve the accuracy and applicability of the estimation.

[0032] In the training process of the material surface reflection distribution function estimation neural network, after the parameters are initialized, a variety of different materials are supervised and learned, so that their parameters reach a relatively optimized stage; in the supervised learning process, the input material image is accompanied by the standard material surface reflection distribution function measurement value as the learning goal, and the neural network training is to minimize the estimation error as the goal, quickly adjust the network parameters, and obtain the material surface reflection distribution function estimation network model. Moreover, after the single material estimation and correction are completed, it also includes the joint comprehensive learning of various materials to construct a total loss function containing various materials: This loss function enables the neural network to learn various materials, including anisotropic materials, so that the learned neural network can accurately estimate the surface reflectance distribution function values ​​of various materials.

[0033] It should be understood that all combinations of the aforementioned concepts and the additional concepts described in more detail below can be considered as part of the inventive subject matter of the present disclosure as long as such concepts are not mutually inconsistent. In addition, all combinations of the claimed subject matter are considered as part of the inventive subject matter of the present disclosure.

[0034] The foregoing and other aspects, embodiments and features of the present invention can be more fully understood from the following description in conjunction with the accompanying drawings. Other additional aspects of the present invention, such as the features and / or beneficial effects of the exemplary embodiments, will be apparent from the following description or learned from the practice of the specific embodiments according to the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] The drawings are not intended to be drawn to scale. In the drawings, each identical or nearly identical component shown in various figures may be represented by the same reference numeral. For clarity, not every component is labeled in every figure. Embodiments of various aspects of the present invention will now be described by way of example and with reference to the accompanying drawings, in which:

[0036] Figure 1 It is a flow chart of a method for measuring bidirectional reflectance distribution function of a digital asset material surface according to an embodiment of the present invention.

[0037] Figure 2A , 2B They are schematic diagrams of material photos of embodiments of the present invention, wherein the red boxes represent overlapping areas. DETAILED DESCRIPTION

[0038] In order to better understand the technical content of the present invention, specific embodiments are given and described as follows in conjunction with the accompanying drawings.

[0039] Various aspects of the present invention are described in this disclosure with reference to the accompanying drawings, in which many illustrative embodiments are shown. The embodiments of the present disclosure are not necessarily intended to include all aspects of the present invention. It should be understood that the various concepts and embodiments introduced above, as well as those described in more detail below, can be implemented in any of many ways, because the concepts and embodiments disclosed by the present invention are not limited to any implementation. In addition, some aspects disclosed by the present invention can be used alone or in any appropriate combination with other aspects disclosed by the present invention.

[0040] Method for measuring bidirectional reflectance distribution function of digital asset material surface

[0041] Combination Figure 1 The exemplary process of the embodiment shown, a method for measuring the bidirectional reflectance distribution function of a digital asset material surface, comprises the following steps:

[0042] Step 1: Use a polarized light source to emit polarized light to the surface of the material to be tested;

[0043] Step 2: Use a polarization camera to capture images of the irradiated surface of the material to be tested, collect the reflection of the light on the surface of the material to be tested, and obtain multiple material photos, wherein the multiple material photos have repeated areas;

[0044] Step 3, extracting repeated areas in multiple material photos;

[0045] Step 4: Based on multiple material photos, estimate the bidirectional reflectance distribution function of each position in the repeated area to obtain p*q bidirectional reflectance distribution functions, where p and q represent the number of position points in the repeated area and the number of material photos, respectively.

[0046] Step 5: Taking the q bidirectional reflectance distribution functions at each position in the repeated area as input, outputting the optimal estimated values ​​of the q bidirectional reflectance distribution functions through the pre-trained neural network prediction model;

[0047] Step 6, obtaining the similarity between the material to be tested and the historical material processed by the neural network prediction model: if the similarity is greater than or equal to the preset threshold, output the optimal estimate obtained in the aforementioned step 5 as the bidirectional reflectance distribution function measurement result of the material to be tested; if the similarity is lower than the preset threshold, proceed to step 7 for a secondary detection process;

[0048] Step 7: Calculate the variance between the optimal estimates of the q bidirectional reflectance distribution functions obtained. If the variance is less than or equal to the preset threshold σ sd , then the optimal estimate obtained in the above step 5 is output as the bidirectional reflectance distribution function measurement result of the material to be tested; if the variance is greater than the preset threshold σ sd , then control the second imaging acquisition of the irradiated surface of the material to be tested and estimate the output through the neural network prediction model, recalculate and judge whether the variance between the optimal estimated values ​​of the bidirectional reflectance distribution function is less than or equal to the preset threshold σ sd , if the variance is less than or equal to the preset threshold σ sd , then the optimal estimated value of the re-estimated output is output as the bidirectional reflectance distribution function measurement result of the material to be measured, otherwise, go to step 8 to calibrate the infrared wave measurement equipment;

[0049] Step 8. Use infrared wave measuring equipment to measure the surface of the material to be tested, calculate the bidirectional reflection distribution function of the material to be tested, and use this as a correction value. Use the p*q bidirectional reflection distribution functions obtained for the first or second time as optional values ​​to fine-tune the neural network prediction model. During the fine-tuning training process, comprehensive learning is performed by combining the total loss functions of all processed materials.

[0050] As an optional embodiment, in the aforementioned steps, the polarized light source and the polarized camera used are configured to have the same polarization state, such as linear polarization or circular polarization. In particular, the polarized light source and the polarized camera used are configured with the same polarizer.

[0051] In the process of scanning and photographing, the polarization camera can be preferably installed on a step-driven mechanical arm. In the process of imaging and collecting the surface of the material to be measured, the position of the polarization camera for each photograph is controlled by a preset program, thereby obtaining the coverage coordinates of the measurement area of ​​each photograph. Therefore, based on the coverage coordinates of the measurement area of ​​each photograph, the repeated areas in multiple material photos are further determined.

[0052] As an optional embodiment, in step 4, estimating the bidirectional reflectance distribution function at each position in the repeated area includes:

[0053] Based on the reflectivity of each position reflected into the direction of the polarized camera lens and the irradiance of the polarized light source irradiating each position, the bidirectional reflectance distribution function is calculated: the reflectivity reflected into the direction of the polarized camera lens / the irradiance of the polarized light source irradiating each position.

[0054] As an optional embodiment, the training process of the pre-trained neural network prediction model includes:

[0055] Initialize a neural network with multiple convolutional layers and fully connected layers;

[0056] The neural network is trained using the collected and calibrated standard material data in the sample database, wherein the standard material data includes optional values ​​of bidirectional reflectance distribution functions of various material objects and calibrated standard values;

[0057] The neural network prediction model is obtained through multiple trainings.

[0058] In an embodiment of the present invention, the loss function of the neural network training process is:

[0059]

[0060] In the formula, e n is the total estimated loss error of the nth material, b is the corrected standard value, is the bidirectional reflectance distribution function measurement value output by the neural network estimation, and m is the position point to be measured in a single material that needs to be estimated.

[0061] In an embodiment of the present invention, obtaining the similarity between the material to be tested and the historical material processed by the neural network prediction model includes:

[0062] Based on the reflectivity of the material to be tested and the reflectivity of the historical material, the similarity between the two is calculated: reflectivity of the material to be tested / reflectivity of the historical material.

[0063] It should be understood that the reflectivity of the material can be obtained by searching the technical manual or related public information in the prior art. The greater the difference in the reflectivity of the material, the lower the similarity, and the lower the confidence of the primary estimation by the neural network. Therefore, in the embodiment of the present invention, a secondary detection is required.

[0064] It should be understood that in the aforementioned step 8, after measuring the surface of the material to be measured by using an infrared wave measuring device and calculating the bidirectional reflection distribution function of the material to be measured and obtaining the BRDF correction value, further fine-tuning training is performed, using the p*q bidirectional reflection distribution functions obtained for the first or second time as optional values, and fine-tuning the neural network prediction model. During the fine-tuning training process, comprehensive learning is performed by combining the total loss functions of all processed materials.

[0065] As an optional method, various materials need to be combined for comprehensive learning during tuning, and the total loss function used is:

[0066]

[0067] Wherein, N represents N different materials.

[0068] Thus, an optimized model after fine-tuning training can be obtained.

[0069] On this basis, since in the solution proposed by the present invention, the neural network is constantly learning and evolving based on the processed materials, especially new materials, after each fine-tuning training, we can make prediction estimates again and re-predict the historical prediction values ​​to obtain more accurate prediction values.

[0070] It can be seen that, through the embodiments of the present invention, in the measurement process, a polarized light source is first used to emit polarized light to the surface of the material to be measured, and a polarized camera is used to capture the reflection of light on the surface of the material to be measured, and multiple material photos are obtained, and the multiple material photos have repeated areas. Then the repeated areas are extracted, and the bidirectional reflectance distribution function of each position, that is, multiple optional values, is obtained, which is used as input, and the optimal estimate of BRDF is output through a pre-trained neural network prediction model, and the similarity between the material to be measured and the historical material is combined to determine whether to perform a secondary test, and the neural network prediction model is fine-tuned and trained again with the optimal estimate or the correction value results of the secondary test and infrared correction, and the total loss function of all processed materials is combined for comprehensive learning, so that the model can be applied to different materials, and high-precision measurement results can be obtained for anisotropic materials.

[0071] Next, we combine the attached Figure 2A , 2B Taking the brick wall material shown as an example, the implementation process of the above embodiment is further explained.

[0072] In order to obtain the surface reflection distribution function of the red brick wall material, we need to measure the light reflection on the surface of the digital asset material to be tested.

[0073] S101: emit polarized light to the surface of an object.

[0074] S102: Use a polarization camera to capture the reflection of light, and collect images from multiple angles through continuous scanning and shooting. There is a lot of overlap in the imaging coverage area.

[0075] The light source and camera are equipped with polarizing filters to ensure that the emitted light source is polarized light. Therefore, after capturing the light signal, the light signal is converted into an electrical signal to obtain a material photo. Figure 2A , 2B shown.

[0076] In order to ensure good measurement results, a certain repetition rate is required when shooting the material surface, that is, a certain amount of overlapping area is required in different photos.

[0077] S103: Mark out the overlapping area.

[0078] Because the camera is fixed on the robot arm and moves during the measurement process, that is, the position of the camera each time it takes a picture is controlled by the program, so the coverage coordinates of the measurement area can be directly obtained each time.

[0079] For any two measurement results, we can directly calculate the overlapping area from their coverage. Figure 2A , 2B In the two images, the parts in the red boxes are overlapping.

[0080] Thus, the overlapping area provides multiple BRDF values ​​to choose from for each position.

[0081] In this example, the BRDF of each position point can be obtained based on existing estimation methods, such as calculating the bidirectional reflectance distribution function based on the reflectivity of each position reflected into the direction of the polarized camera lens and the irradiance of the polarized light source irradiating each position: the reflectivity reflected into the direction of the polarized camera lens / irradiance of the polarized light source irradiating each position.

[0082] S104: Based on the trained neural network prediction model, the optional value of the BRDF is input and the optimal value of the output BRDF can be estimated.

[0083] In this embodiment, we propose a neural network that inputs multiple optional values ​​x at a certain position of material M m1 , x m2 , x m3 , ..., x mn , the neural network outputs an optimal value

[0084] The neural network responsible for estimating the optimal value is pre-trained, for example, using a neural network architecture that includes multiple convolutional layers and fully connected layers. After the neural network is randomly initialized, we use the standard material data that has been collected and calibrated in the sample database as training data to train the neural network.

[0085] Standard material data includes optional BRDF values ​​for various material objects and calibrated standard values.

[0086] In this example, we use the high-precision native scanned PBR digital asset library PBRMAX ( www.pbrmax.cn / ? activeIndex=0 ) as training data to train the neural network.

[0087] The training loss function is where e n is the total estimated loss error of the nth material, b is the measured value of the standard material surface reflection distribution function, is the measured value estimated by the neural network, and m is the position point to be measured that needs to be estimated in a single material.

[0088] The training process of the neural network aims to minimize the estimation error and quickly adjust the network parameters to complete the training.

[0089] After training, the neural network model can be used to predict the optimal BRDF value based on the optional BRDF values ​​of the overlapping area of ​​the new material as input.

[0090] In the embodiment of the present invention, the aforementioned pre-trained neural network model will continuously learn and optimize autonomously while predicting new materials during use. Therefore, it is expected that the neural network model can be applicable to various materials rather than a single material.

[0091] Therefore, in an optional embodiment, after predicting the new material, we use the prediction result of the new material to perform fine-tuning training to obtain an updated neural network model.

[0092] In the process of predicting and estimating new materials, the prediction results may be inaccurate due to the lack of calibration values ​​for new materials. Therefore, we analyze the similarity between each new material and the historical material. For example, we estimate it based on the similarity of reflectivity.

[0093] Based on the reflectivity of the material n1 to be tested and the reflectivity of the historical material n2, calculate the similarity s between the two n1，n2 : Reflectivity of the material to be tested / reflectivity of the historical material.

[0094] The greater the difference in reflectivity of the materials, the lower the similarity, and the confidence of the first estimation by the neural network is considered to be relatively low. Therefore, in the embodiment of the present invention, a second detection is required.

[0095] That is, if the similarity is lower than a certain threshold σ s We believe that the new material is too different from the historical material, and the reliability of the estimated value of the estimation network is easily weakened accordingly. The predicted value of the estimation network lacks credibility and needs to be retested.

[0096] The secondary test calculates the variance between each estimate and the estimated value. If the variance is greater than a certain threshold σ sd , then the area is re-sampled until the variance is lower than the threshold σ sd .

[0097] If the error cannot be eliminated after the second acquisition, the near-infrared wave device is called for correction. After the second detection and correction, after obtaining the BRDF correction value, we also fine-tune the neural network model according to the BRDF correction value to achieve the purpose of lifelong learning.

[0098] It should be understood that in each fine-tuning training, various materials need to be combined for comprehensive learning, and the total loss function is:

[0099] Where N represents all different materials.

[0100] In this way, the fine-tuning training process is completed with the goal of minimizing the total estimation error, achieving lifelong learning and improving the accuracy of estimation of a variety of materials.

[0101] During the fine-tuning training process, China Lianli conducts comprehensive learning of various materials, so that the neural network can learn various materials, including anisotropic materials. Therefore, the learned neural network can accurately estimate the surface reflection distribution function values ​​of various materials.

[0102] Computer readable medium

[0103] According to the embodiments disclosed in the present invention, a computer-readable medium storing software is also proposed, wherein the software includes instructions that can be executed by one or more computers, and the aforementioned instructions, through such execution, enable the one or more computers to perform operations, and these operations include the process of the method of any of the aforementioned embodiments.

[0104] Computer Systems

[0105] According to an embodiment disclosed in the present invention, a computer system is further provided, comprising: one or more processors and at least one memory.

[0106] The memory stores operable instructions, and the aforementioned instructions, through execution, enable the one or more computers to perform operations, and these operations include the process of the method of any of the aforementioned embodiments.

[0107] Although the present invention has been disclosed as above with preferred embodiments, it is not intended to limit the present invention. A person with ordinary knowledge in the technical field to which the present invention belongs may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the protection scope of the present invention shall be determined by the definition of the claims.

Claims

1. A method for measuring the bidirectional reflectance distribution function of a digital asset material surface, characterized in that: The following steps are involved: Step 1: Use a polarized light source to emit polarized light to the surface of the material to be tested; Step 2: Use a polarization camera to capture images of the irradiated surface of the material to be tested, collect the reflection of the light on the surface of the material to be tested, and obtain multiple material photos, wherein the multiple material photos have repeated areas; Step 3, extracting repeated areas in multiple material photos; Step 4: Based on multiple material photos, estimate the bidirectional reflectance distribution function of each position in the repeated area to obtain p*q bidirectional reflectance distribution functions, where p and q represent the number of position points in the repeated area and the number of material photos, respectively. Step 5: Taking the q bidirectional reflectance distribution functions at each position in the repeated area as input, outputting the optimal estimated values ​​of the q bidirectional reflectance distribution functions through the pre-trained neural network prediction model; Step 6, obtaining the similarity between the material to be tested and the historical material processed by the neural network prediction model: if the similarity is greater than or equal to the preset threshold, output the optimal estimate obtained in the aforementioned step 5 as the bidirectional reflectance distribution function measurement result of the material to be tested; if the similarity is lower than the preset threshold, proceed to step 7 for a secondary detection process; Step 7: Calculate the variance between the optimal estimates of the q bidirectional reflectance distribution functions obtained. If the variance is less than or equal to the preset threshold σ sd , then the optimal estimate obtained in the above step 5 is output as the bidirectional reflectance distribution function measurement result of the material to be tested; if the variance is greater than the preset threshold σ sd , then control the second imaging acquisition of the irradiated surface of the material to be tested and estimate the output through the neural network prediction model, recalculate and judge whether the variance between the optimal estimated values ​​of the bidirectional reflectance distribution function is less than or equal to the preset threshold σ sd , if the variance is less than or equal to the preset threshold σ sd , then the optimal estimated value of the re-estimated output is output as the bidirectional reflectance distribution function measurement result of the material to be measured, otherwise, go to step 8 to calibrate the infrared wave measurement equipment; Step 8. Use infrared wave measuring equipment to measure the surface of the material to be tested, calculate the bidirectional reflection distribution function of the material to be tested, and use this as a correction value. Use the p*q bidirectional reflection distribution functions obtained for the first or second time as optional values ​​to fine-tune the neural network prediction model. During the fine-tuning training process, comprehensive learning is performed by combining the total loss functions of all processed materials.

2. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The polarized light source used and the polarization camera are configured to have the same polarization state.

3. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The polarized light source and polarized camera used are both equipped with the same polarizing filter.

4. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The polarization camera is installed on a step-driven mechanical arm. During the process of imaging and collecting the surface of the material to be measured, the position of the polarization camera for each photo is controlled by a preset program, thereby obtaining the coverage range coordinates of the measurement area for each photo; Based on the coverage coordinates of the measurement area of ​​each photo, the repeated areas in multiple material photos are determined.

5. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: In step 4, the estimation of the bidirectional reflectance distribution function at each position in the repeated area includes: Based on the reflectivity of each position reflected into the direction of the polarized camera lens and the irradiance of the polarized light source irradiating each position, the bidirectional reflectance distribution function is calculated: the reflectivity reflected into the direction of the polarized camera lens / the irradiance of the polarized light source irradiating each position.

6. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The training process of the pre-trained neural network prediction model includes: Initialize a neural network with multiple convolutional layers and fully connected layers; The neural network is trained using the collected and calibrated standard material data in the sample database, wherein the standard material data includes optional values ​​of bidirectional reflectance distribution functions of various material objects and calibrated standard values; The neural network prediction model is obtained through multiple trainings, wherein the loss function of the training process is: In the formula, e n is the total estimated loss error of the nth material, b is the corrected standard value, is the bidirectional reflectance distribution function measurement value output by the neural network estimation, and m is the position point to be measured in a single material that needs to be estimated.

7. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The total loss function is: Wherein, N represents N different materials.

8. The method for measuring bidirectional reflectance distribution function of digital asset material surface according to claim 1, characterized in that: The obtaining of the similarity between the material to be tested and the historical material processed by the neural network prediction model includes: Based on the reflectivity of the material to be tested and the reflectivity of the historical material, the similarity between the two is calculated: reflectivity of the material to be tested / reflectivity of the historical material.

9. A computer-readable medium storing software, characterized in that: The software includes instructions that can be executed by one or more computers, and the instructions, through such execution, enable the one or more computers to perform operations, wherein the operations include the process of the method according to any one of claims 1 to 8.

10. A computer system, characterized in that: include: one or more processors; A memory storing operable instructions, wherein when the instructions are executed by the one or more processors, the one or more processors are caused to perform operations, wherein the operations include the process of the method according to any one of claims 1 to 8.

Images