Training methods, devices, electronic equipment, and storage media for depth estimation networks

Abstract

This application provides a training method, apparatus, electronic device, and storage medium for a depth estimation network. The method includes acquiring a first image and a second image of a target object, and acquiring a first depth image of the first image; acquiring a pixel pose transformation relationship between the first image and the second image, wherein the pixel pose transformation relationship includes the positional relationship between each first pixel in the first image and a second pixel in the second image, wherein the first pixel and the second pixel correspond to the same part of the target object; generating a reconstructed image based on the first depth image, the pixel pose transformation relationship, and pre-acquired camera parameters; determining the loss of the depth estimation network based on the differences between the first image, the depth image, the reconstructed image, and the second image; and adjusting the parameters of the depth estimation network based on the loss of the depth estimation network. This improves the accuracy of pose transformation of moving objects.

Description

Technical Field

[0001] This invention relates to the field of computer vision technology, and more specifically to a training method, apparatus, electronic device, and storage medium for a depth estimation network. Background Technology

[0002] In the field of computer vision, pose estimation is an important topic. The PoseNet algorithm model uses a convolutional neural network to predict the 6-DOF pose of a camera. However, this pose network is used to output the camera pose transformation matrix, which means that this pose transformation matrix can only produce correct results for static objects. Using this pose transformation matrix for moving objects will introduce errors. Summary of the Invention

[0003] In view of the above, it is necessary to propose a training method, device, electronic device and storage medium for a depth estimation network, which can solve the problem that the transformation matrix of a moving object is inconsistent with the camera transformation matrix when there is a moving object in the image, and improve the accuracy of the pose transformation of the moving object.

[0004] In a first aspect, embodiments of this application provide a training method for a depth estimation network. The method includes acquiring a first image and a second image of a target object, wherein the first image and the second image are images taken at different times; inputting the first image into a depth estimation network to obtain a first depth image; inputting the first image and the second image into the depth estimation network to obtain a pixel pose transformation relationship between the first image and the second image, wherein the pixel pose transformation relationship includes the positional relationship between a first pixel in the first image and a second pixel in the second image, wherein the first pixel and the second pixel both correspond to the same part of the target object; generating a reconstructed image based on the first depth image, the pixel pose transformation relationship, and pre-acquired camera parameters; determining the loss of the depth estimation network based on the differences between the first image, the depth image, the reconstructed image, and the second image; and adjusting the parameters of the depth estimation network based on the loss of the depth estimation network.

[0005] Secondly, embodiments of this application provide a training apparatus for a depth estimation network, comprising: an image acquisition module for acquiring a first image and a second image of a target object, wherein the first image and the second image are images taken at different times; a depth acquisition module for inputting the first image into the depth estimation network to obtain a first depth image; a pose acquisition module for inputting the first image and the second image into the depth estimation network to obtain a pixel pose transformation relationship between the first image and the second image, wherein the pixel pose transformation relationship includes the positional relationship between a first pixel on the first image and a second pixel on the second image, wherein the first pixel and the second pixel correspond to the same part of the target object; a generation module for generating a reconstructed image based on the first depth image, the pixel pose transformation relationship, and pre-acquired camera parameters; a determination module for determining the loss of the depth estimation network based on the differences between the first image, the depth image, the reconstructed image, and the second image; and an adjustment module for adjusting the parameters of the depth estimation network based on the loss of the depth estimation network.

[0006] Thirdly, embodiments of this application provide an electronic device, the electronic device including a memory and a processor, the memory being used to store at least one instruction, and the processor being used to implement the training method of the depth estimation network as described above when executing the at least one instruction.

[0007] Fourthly, embodiments of this application provide a computer-readable storage medium storing at least one instruction that, when executed by a processor, implements the training method for the depth estimation network as described above.

[0008] Compared to existing technologies, the depth estimation network training method, apparatus, electronic device, and storage medium provided by this invention address the issue that existing pose transformation networks output camera pose transformation matrices, which only yield correct results for static objects. This application's embodiments, by outputting the pixel pose transformation relationship for each pixel, solve the problem of inconsistency between the transformation matrix of a moving object and the camera's transformation matrix in images, thus improving the accuracy of pose transformation for moving objects. Attached Figure Description

[0009] Figure 1 This is a schematic diagram of an electronic device structure provided in an embodiment of this application.

[0010] Figure 2 A schematic diagram of the training device structure for the depth estimation network provided in this application embodiment.

[0011] Figure 3This is a schematic diagram of a training method for a depth estimation network provided in an embodiment of this application.

[0012] Figure 4 This is a schematic diagram of another training method for a depth estimation network provided in an embodiment of this application.

[0013] The following detailed description, in conjunction with the accompanying drawings, will further illustrate the present invention. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Before providing a detailed explanation of the embodiments of this application, the application scenarios involved in the embodiments of this application will be introduced first.

[0016] The transformation matrix (RT) obtained by the existing PoseNet algorithm model is the pose change of the camera from time t to t+1, i.e., the camera transformation matrix (RT). If there is a moving object in the image, the transformation matrix of the moving object is inconsistent with the camera transformation matrix, resulting in the inability to obtain the accurate pose change of the moving object. To address this, the training method, apparatus, electronic device, and storage medium for the depth estimation network provided in this application obtain the pixel pose transformation matrix of each pixel, allowing each pixel to learn and adjust its corresponding pose transformation based on its corresponding pixel pose transformation matrix. This solves the problem of inconsistency between the transformation matrix of a moving object and the camera transformation matrix when there is a moving object in the image, thus improving the accuracy of the pose transformation of the moving object.

[0017] The training method, apparatus, electronic device, and storage medium for the depth estimation network provided in this application embodiment can be applied in fields such as autonomous driving, scene understanding, robotics, 3D reconstruction, photography and videography, intelligent medicine, intelligent human-computer interaction, spatial mapping, and augmented reality.

[0018] Please see Figure 1 This application provides an embodiment of an electronic device 100. It includes a memory 10, a processor 20, and a training device 200 for a depth estimation network stored in the memory 10 and executable on the processor 20. When the processor 20 executes the training device 200 for the depth estimation network, it implements the steps in the training method embodiment of the depth estimation network, for example... Figures 3 to 4The steps shown. Alternatively, when the processor 20 executes the training apparatus 200 for the depth estimation network, it implements the functions of each module in the network training program embodiment, for example... Figure 2 Modules 210-260 in the middle.

[0019] The training device 200 for the deep estimation network can be divided into one or more modules. One or more modules are stored in the memory 10 and executed by the processor 20 to complete the embodiments of this application. The one or more modules can be a series of computer program instruction segments capable of performing specific functions. These instruction segments describe the execution process of the training device 200 for the deep estimation network in the electronic device 100. For example, the training device 200 for the deep estimation network can be divided into... Figure 2 The system includes an image acquisition module 210, a depth acquisition module 220, a pose acquisition module 230, a generation module 240, a determination module 250, and an adjustment module 260. For the specific functions of each module, please refer to the descriptions below.

[0020] The processor 20 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor, and the processor 20 may be connected to various parts of the electronic device 100 using various interfaces and buses.

[0021] The memory 10 can be used to store the training device 200 and / or modules of the depth estimation network. The processor 20 implements various functions of the electronic device 100 by running or executing computer programs and / or modules stored in the memory 10 and by calling data stored in the memory 10. The memory 10 may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0022] In one embodiment, the electronic device 100 further includes a communication unit 30, which is used to establish a communication connection with other computer devices via wired or wireless means. The communication unit 30 may be a wired communication unit or a wireless communication unit.

[0023] The electronic device 100 may also include an input / output unit 40, which includes a keyboard, mouse, display screen, etc. The display screen is used to display media files of the electronic device 100.

[0024] Electronic device 100 may be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. Those skilled in the art will understand that the schematic diagram is merely an example of electronic device 100 and does not constitute a limitation on electronic device 100. It may include more or fewer components than shown, or combine certain components, or different components. For example, electronic device 100 may also include network access devices, buses, etc.

[0025] Please see Figure 2 This application provides an embodiment of the structure of a training device 200 for a depth estimation network.

[0026] See Figure 2 As shown, the training device 200 for the depth estimation network may include an image acquisition module 210, a depth acquisition module 220, a pose acquisition module 230, a generation module 240, a determination module 250, and an adjustment module 260. In one embodiment, the above modules may be programmable software instructions stored in the memory 10 and executable by the processor 20. It is understood that in other embodiments, the above modules may also be program instructions or firmware embedded in the processor 20.

[0027] The image acquisition module 210 is used to acquire a first image and a second image of the target object, wherein the first image and the second image are images at different times.

[0028] The depth acquisition module 220 is used to input the first image into the depth estimation network to obtain a first depth image.

[0029] The pose acquisition module 230 is used to input the first image and the second image into the depth estimation network to obtain the pixel pose transformation relationship between the first image and the second image. The pixel pose transformation relationship includes the positional relationship between the first pixel on the first image and the second pixel on the second image for each first pixel on the first image, wherein the first pixel and the second pixel both correspond to the same part of the target object.

[0030] The generation module 240 is used to generate a reconstructed image based on the first depth image, the pixel pose transformation relationship, and the pre-acquired camera parameters.

[0031] The determination module 250 is used to determine the loss of the depth estimation network based on the differences between the first image, the depth image, the reconstructed image and the second image.

[0032] The adjustment module 260 is used to adjust the parameters of the depth estimation network based on the loss of the depth estimation network.

[0033] Figure 3 This is a flowchart illustrating the training method of the depth estimation network in Embodiment 1 of this application. The order of steps in the flowchart can be changed, and some steps can be omitted, depending on different requirements.

[0034] Step S30: Obtain a first image and a second image of the target object, wherein the first image and the second image are images taken at different times.

[0035] In this embodiment of the application, training the depth estimation network requires acquiring training image data, which includes a first image and a second image. Both the first and second images are images of the same target object, but were taken at different times. For example, the first image is the image at time t, and the second image is the image at time t+1. The target object can be, for example, a person, an animal, a vehicle, or a building. Both the first and second images include movable objects, such as vehicles or people moving relative to the ground.

[0036] Specifically, a training set of images of the vehicle's movement is established using a monocular camera. This means a monocular camera is mounted on the vehicle and continuously captures images. The images captured by the monocular camera constitute the training set. From this training set, RGB images at times t and t+1 are randomly sampled. t I t+1 , where I t The subscript t represents time t, I t+1 The subscript t+1 represents time t+1.

[0037] In this embodiment of the application, the depth estimation network includes a depth prediction subnetwork and a pose estimation subnetwork.

[0038] Deep prediction subnetworks can be autoencoder networks. Autoencoders are a class of artificial neural networks (ANNs) used in semi-supervised and unsupervised learning. Their function is to learn representations of input information by using the input information as the learning target. An autoencoder consists of an encoder and a decoder. Based on learning paradigm, autoencoders can be divided into contractive autoencoders, regularized autoencoders, and variational autoencoders (VAEs), where the former two are discriminative models and the latter is a generative model. Based on architecture, autoencoders can be feedforward or recursive neural networks.

[0039] In some embodiments, the depth prediction subnetwork includes a depth encoder and a depth decoder. The depth estimation model feeds a first image through the depth encoder to extract abstract depth features. The depth encoder can use different machine learning algorithms and techniques to extract these features.

[0040] In some embodiments, the pose estimation subnetwork is a deep convolutional neural network.

[0041] Step S31: Input the first image into the depth estimation network to obtain the first depth image.

[0042] In this embodiment, step S31 can be implemented by inputting the first image into a depth estimation subnetwork and outputting a first depth image based on the parameters of the depth estimation subnetwork. In this embodiment, the first image is input into a depth estimation model, which outputs a first depth image corresponding to the first image. The pixel value (or grayscale value) of each pixel in the first depth image can represent either relative depth or depth value. Relative depth can be understood as the logical relationship between the preceding and following pixels. The distance between the real object corresponding to a pixel and the camera device or reference plane is called the depth value of that pixel, that is, the depth value of a pixel is the vertical distance from the real object corresponding to that point to the aforementioned camera device.

[0043] For example, a deep encoder is a convolutional neural network consisting of multiple layers, where each subsequent layer reduces the dimensionality of the extracted features. For instance, after the first layer, the dimensionality would be on the order of 10. ^ An input image of 6 pixels or data points is reduced to an order of magnitude 10. ^ A set of features of 5. Through the last layer in the deep encoder, abstract deep features can be obtained in 10. ^4 or smaller. These numbers are purely for illustrative purposes. In reality, depth encoders can have different numbers of layers, and the number of pixels and depth features can vary.

[0044] In contrast, deep decoders include multiple layers to increase the dimensionality of abstract features.

[0045] As in the example above, the depth decoder can be on the order of 10. ^ The depth features are abstracted from 4, and the depth (relative depth or depth value) at each pixel of the input image is progressively derived over multiple layers. Then, the depth decoder outputs a first depth image, where each pixel on the depth map corresponds to the distance of the nearest object in the scene projected to that pixel.

[0046] In some embodiments, the depth decoder outputs a disparity map, where each pixel on the disparity map corresponds to the reciprocal of the distance. In embodiments of this application, the principles described with reference to the depth map are readily applied to implementations with disparity maps. For example, a first image has already captured a tree at a given pixel at an unknown distance from the camera. The depth decoder outputs a depth value corresponding to the distance from the camera to the block at that pixel.

[0047] In some embodiments, the output depth value may be relative to another depth value. In other embodiments, the output depth value is scaled to provide a realistic measurement of objects in the scene, such as a city block being 10 meters away or a building being 25 meters away.

[0048] Step S32: Input the first image and the second image into the depth estimation network to obtain the pixel pose transformation relationship between the first image and the second image. The pixel pose transformation relationship includes the positional relationship between the first pixel on the first image and the second pixel on the second image for each first pixel on the first image, wherein the first pixel and the second pixel correspond to the same part of the target object.

[0049] In this embodiment, step S32 can be implemented by inputting the first image and the second image into the pose estimation subnetwork to extract the pixel pose transformation relationship between the first image and the second image. That is, the pixel pose transformation relationship output by the pose estimation subnetwork includes the pixel transformation matrix of all pixels in the first image or the second image, and the pixel transformation matrix is ​​the pose transformation relationship of a certain pixel in the first image and the second image.

[0050] In embodiments of this application, a pose estimation subnetwork receives multiple cascaded images (such as a first image and a second image) to determine the pose of each input image. Pose generally refers to a mathematical transformation between perspectives of two images. In some embodiments, pose more generally describes the perspective of an image, where the perspective can be used to define the transformation between two images.

[0051] In this embodiment, the pose estimation subnetwork may include a pose encoder and a pose decoy. A first image and a second image are fed into the pose encoder, which extracts abstract pose features from the first and second images. These abstract pose features are then input to a pose decoder, which determines the pixel pose transformation relationship between the first and second images. The pose encoder may be configured as a convolutional neural network comprising multiple layers to extract the abstract pose features and then derive the pose of the first and second images.

[0052] For example, for point P on the target object, which corresponds to a first pixel P1 in the first image and a second pixel P2 in the second image, the pose estimation subnetwork solves for the pixel pose transformation relationship between the first pixel P1 and the second pixel P2. Similarly, the pixel pose transformation relationship for all pixels in the first image is solved.

[0053] Step S33: Generate a reconstructed image based on the first depth image, pixel pose transformation relationship, and pre-acquired camera parameters.

[0054] In this embodiment of the application, after acquiring the first depth image, a reconstructed image can be generated based on the depth values ​​of each pixel in the first depth image, the pixel pose transformation relationship between the first image and the second image, and the camera parameters.

[0055] Specifically, the depth value of each pixel in the first depth image, the pre-acquired camera parameters, the pixel pose transformation relationship, and the position coordinates of each pixel in the first image are used as inputs to the position transformation algorithm to obtain the position coordinates of each pixel in the reconstructed image; the pixel value of each pixel in the first depth image is determined, and the reconstructed image is generated based on the pixel value and position coordinates.

[0056] Specifically, the formula for the position transformation algorithm is shown in Equation (1). By inputting the depth value of each pixel in the first depth image, camera parameters, pixel pose transformation relationship, and position coordinates of each pixel in the first image into Equation (1), the position of each pixel in the first image in the reconstructed image can be obtained.

[0057] B2 = KTD1K -1 Formula B1 (1)

[0058] Where B1 is the position of each pixel in the first image, represented by (x1, y1), B2 is the position of each pixel in the first image in the reconstructed image, represented by (x2, y2), K is a matrix composed of camera parameters, and T is the pixel pose transformation relationship. The number of columns in K is equal to the number of rows in T.

[0059] Step S34: Determine the loss of the depth estimation network based on the differences between the first image, the depth image, the reconstructed image, and the second image.

[0060] Please refer to the following: Figure 4 , Figure 4 This is a schematic diagram of a training method for a depth estimation network provided in an embodiment of this application.

[0061] Step S41: Calculate the brightness error between the first image and the reconstructed image using the first loss function.

[0062] Step S42: Calculate the gradient error between the first image and the first depth image using the second loss function.

[0063] Step S43: Calculate the loss of the depth estimation network based on the brightness error and the gradient error.

[0064] The first loss function is:

[0065]

[0066] in, This indicates that the target object is in the first image I. t The gradient of the region in which it is located, This indicates that the target object is in the first depth image D. t The gradient of the region in which the target object is located, S is the gradient of the region in the first image I. t The set of pixels in the region in which it is located;

[0067] The second loss function is:

[0068] L ph (I t ,I` t )=(1-α)|I` t (u, v)-I t (u, v)|+αL SSIM ;

[0069] Among them, L ph (I t ,I` t ) represents the brightness consistency error, I t Represents the reconstructed image, I` t This represents the second image, where α is a weighted balancing hyperparameter, and L... SSIM =(1-SSIM(I) t ,I` t () / 2), SSIM is a metric used to measure the similarity between the reconstructed image and the second image, |I` t(u, v)-I t (u, v)| represents the grayscale difference between the reconstructed image and the second image.

[0070] Step S35: Adjust the parameters of the depth estimation network based on the loss of the depth estimation network.

[0071] In this embodiment of the application, step S35 can be implemented by iteratively updating the parameters of the pose estimation subnetwork and the depth estimation subnetwork by minimizing the brightness error and the gradient error.

[0072] In this embodiment, the pose estimation subnetwork outputs the pose transformation relationship of each pixel, and then the pose estimation subnetwork learns and adjusts the pose transformation relationship of each pixel. Therefore, it can solve the problem that the transformation matrix of the moving object is inconsistent with the camera transformation matrix when there is a moving object in the image, and improve the accuracy of the pose transformation of the moving object.

[0073] This application also provides a computer-readable storage medium storing at least one instruction that, when executed by a processor, implements the training method for the depth estimation network described above.

[0074] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims. Furthermore, it is clear that the word "comprising" does not exclude other elements or, and the singular does not exclude the plural. Multiple elements or devices recited in the apparatus claims may also be implemented by a single element or device in software or hardware. The terms "first," "second," etc., are used to indicate names and do not indicate any particular order.

[0075] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A method for training a depth estimation network, characterized in that, The method comprises: obtaining a first image and a second image of a target object, wherein the first image and the second image are images at different time points respectively; inputting the first image into a depth estimation network to obtain a first depth image; inputting the first image and the second image into the depth estimation network to obtain a pixel pose transformation relationship between the first image and the second image, wherein the pixel pose transformation relationship comprises a positional relationship between a first pixel on the first image and a second pixel on the second image for each first pixel on the first image, wherein the first pixel and the second pixel both correspond to the same part of the target object; generating a reconstructed image according to the first depth image, the pixel pose transformation relationship and pre-obtained camera parameters; determining a loss of the depth estimation network according to differences between the first image, the depth image, the reconstructed image and the second image; adjusting parameters of the depth estimation network based on the loss of the depth estimation network.

2. The method of claim 1, wherein, The depth estimation network comprises a pose estimation subnetwork; and inputting the first image and the second image into the depth estimation network to obtain the pixel pose transformation relationship between the first image and the second image comprises: inputting the first image and the second image into the pose estimation subnetwork to extract the pixel pose transformation relationship between the first image and the second image.

3. The method of claim 2, wherein, The depth estimation network further comprises a depth estimation subnetwork; and inputting the first image into the depth estimation network to obtain the first depth image comprises: inputting the first image into the depth estimation subnetwork to output the first depth image based on parameters of the depth estimation subnetwork.

4. The method of training a depth estimation network according to any one of claims 1 to 3, wherein, The determination of the loss of the depth estimation network according to the differences between the first image, the depth image, the reconstructed image and the second image comprises: calculating a photometric error between the first image and the reconstructed image using a first loss function; calculating a gradient error between the first image and the first depth image using a second loss function; and calculating the loss of the depth estimation network based on the photometric error and the gradient error.

5. The method of claim 4, wherein, The first loss function is: wherein, represents a gradient of a region in which the target object is located in the first image I t , represents a gradient of a region in which the target object is located in the first depth image D t , S is a set of pixel points of a region in which the target object is located in the first image I t . The second loss function is: ； wherein Lph(I t , I t ) represents a luminosity consistency error, I t represents the reconstructed image, I t represents the second image, a is a weight balancing hyperparameter, LSSIM = (1 - SSIM(I t , I t ) / 2), SSIM is an index for measuring the similarity of the reconstructed image and the second image, and |I t (u, v) - I t (u, v) | represents the grayscale difference between the reconstructed image and the second image.

6. The method of training a depth estimation network of claim 5, wherein, The depth estimation network comprises a pose estimation subnetwork and a depth estimation subnetwork, and adjusting the parameters of the depth estimation network based on the loss of the depth estimation network comprises: iteratively updating the parameters of the pose estimation subnetwork and the depth estimation subnetwork by minimizing the photometric error and the gradient error.

7. The method of training a depth estimation network of claim 6, wherein, The depth estimation subnetwork is a self-encoding network comprising an encoder and a decoder, and the pose estimation subnetwork is a deep convolutional neural network.

8. An apparatus for training a depth estimation network, comprising: The method comprises: an image obtaining module configured to obtain a first image and a second image of a target object, wherein the first image and the second image are images at different time points respectively; a depth obtaining module configured to input the first image into a depth estimation network to obtain a first depth image; The pose obtaining module is configured to input the first image and the second image into the depth estimation network to obtain a pixel pose conversion relationship between the first image and the second image, wherein the pixel pose conversion relationship comprises a positional relationship between each first pixel on the first image and a second pixel on the second image, and the first pixel and the second pixel both correspond to a same part of the target object. The generating module is configured to generate a reconstructed image according to the first depth image, the pixel pose conversion relationship and pre-obtained camera parameters. The determining module is configured to determine a loss of the depth estimation network according to a difference between the first image, the depth image, the reconstructed image and the second image. The adjusting module is configured to adjust parameters of the depth estimation network based on the loss of the depth estimation network.

9. An electronic device, comprising: The electronic device comprises a memory and a processor, the memory is configured to store at least one instruction, and the processor is configured to execute the at least one instruction to implement the training method of the depth estimation network according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, and the at least one instruction is executed by the processor to implement the training method of the depth estimation network according to any one of claims 1 to 7.

Images