Abstract: Methods and tiling engines for tiling primitives in a tile based graphics processing system in which a rendering space is divided into a plurality of tiles. The method includes generating a multi-level hierarchy of tile groups, each level of the multi-level hierarchy comprising one or more tile groups comprising one or more of the plurality of tiles; receiving a plurality of primitive blocks, each primitive block comprising geometry data for one or more primitives; associating each of the plurality of primitive blocks with one or more of the tile groups up to a maximum number of tile groups such that if at least one primitive of a primitive block falls, at least partially, within the bounds of a tile, the primitive block is associated with at least one tile group that includes that tile; and generating a control stream for each tile group based on the associations, wherein each control stream comprises a primitive block entry for each primitive block associated with the corresponding tile group.

Abstract: A method of skin microstructure texture filtering for facial animation includes obtaining a plurality of one-dimensional (1D) filtered tiles corresponding to a plurality of filter axis angles and a plurality of filter parameters applied to a neutral tile, and at runtime, for each pixel representing a region of human skin, determining a principal direction of deformation, a principal filter parameter corresponding to the principal direction of deformation, and a secondary filter parameter corresponding to a secondary direction of deformation orthogonal to the principal direction of deformation, and selecting a first 1D filtered tile among the plurality of 1D filtered tiles, the first 1D filter tile corresponding to the secondary direction of deformation and the secondary filter parameter, and generating a respective two-dimensional (2D) filtered tile by convolving the first 1D filtered tile with a second 1D filter kernel corresponding to the principal direction of deformation and the principal filter parameter

Abstract: Systems and methods to implement a technique for determining an environment importance sampling function. An environment map may be provided where lighting information about the environment is known, but where certain pixels within a scene associated with the environment map are shaded. From these shaded pixels, rays may be drawn in random directions to determine whether the rays are occluded or can interact with the environment map, which provides an indication of a source of lighting that can be used for light transport simulations. A mask may be generated based on these occlusions and used to update the environment importance sampling function.

Abstract: A computer system having a processor executing a set of instructions that cause the processor to receive first image data having pixel information of a geographic region; calculate for particular real-world locations within the geographic region, a plurality of vegetation indices with combinations of the pixel information; generate a land cover mask with the vegetation indices, the land cover mask identifying first real-world locations within the geographic region having a water-related land cover type, a non-vegetated land cover type and an evergreen land cover type; classify second real-world locations within the geographic region that are not classified as the water-related land cover type, the non-vegetated land cover type and the evergreen land cover type as cropland; and analyze a time-series of image data depicting the second real-world locations within the geographic region with phenology metrics to identify at least one particular type of cropland within the second real-world locations.

Abstract: A ray (e.g., a traced path of light, etc.) is generated from an originating pixel within a scene being rendered. Additionally, one or more shadow map lookups are performed for the originating pixel to estimate an intersection of the ray with alpha-tested geometry within the scene. A shadow map stores the distance of geometry as seen from the point of view of the light, and alpha-tested geometry includes objects within the scene being rendered that have a determined texture and opacity. Further, the one or more shadow map lookups are performed to determine a visibility value for the pixel (e.g., that identifies whether the originating pixel is in a shadow) and a distance value for the pixel (e.g., that identifies how far the pixel is from the light). Further still, the visibility value and the distance value for the pixel are passed to a denoiser.

Abstract: Systems and methods are provided for generating virtual three-dimensional environments that allow for simultaneous and collaborate interaction amongst users that utilize virtual reality systems, augmented reality systems, and standard desktop systems. As users interact with elements presented within these virtual three-dimensional environments, these interactions and changes to these elements are propagated to other users within these environments in real time.

Abstract: An image processing apparatus includes a memory storing instructions, and a processor configured to execute the instructions to acquire first image information acquired by imaging using a polarization element configured to transmit lights with a plurality of polarization azimuths that are different from each other, determine whether a luminance of at least part of the first image information becomes higher than a threshold, and acquire polarization information using the first image information and a determination result. A predetermined condition is satisfied.

Abstract: An automatic identification method, apparatus and device for detection results of an immunoreagent card.

Abstract: Systems, methods, and techniques dynamically utilize load balancing for workgroup assignments between a group of shader engines by a command processor of a graphics processing unit (GPU). Based on one or more commands received for execution, a plurality of workgroups is generated for assignment to a plurality of shader engines for processing, each shader engine including a respective quantity of active compute units. Each workgroup of the plurality of workgroups is dynamically assigned to a respective shader engine for execution based at least in part on indications of available resources respectively associated with each of the shader engines. In various embodiments, the indications of available resources may include physical parameters regarding each shader engine, as well as current status information regarding the processing of workgroups assigned to each shader engine.

Abstract: A mesh model of a 3D space is provided with improved accuracy by refining the locations of edges of objects in the space. The mesh model includes vertices which define surfaces of triangles. Triangles are identified which have two vertices in one plane and another, outlier vertex in another, adjacent plane. A line is fitted to the outlier vertices to define an edge of an object, and the outlier vertices are moved to the line, referred to as a mesh-based line. Texture data from images of the space can be used to further refine the edge. In one approach, gradients in grayscale pixels which correspond the vertices of the mesh-based line are used to define a grayscale-based line. The two line definitions can be combined or otherwise used to provide a final definition of the edge. The object can be measured based on the length and position of the edge.

Abstract: Techniques are disclosed relating to distributing geometry work in a graphics processor. In some embodiments, multiple geometry pipelines are configured to process, in parallel, multiple segments of a set of geometry work. In some embodiments, control circuitry is configured to split operations from the set of graphics work into the multiple segments. This may include parsing a control stream for the set of graphics work to determine an initial work estimate based on numbers of primitives in different draw calls and adjusting the initial work estimate based on a complexity determination for the draw call.

Abstract: A system and method builds triangles from vertices of polygons and stores the pixels overlayed at least in part by each triangle. The system and method then takes pixels overlayed by an even number of times for each polygon and renders them. Antialiasing information may be generated and provided by oring or exclusive oring bits representing each polygon for each of several pixel divisions.

Abstract: Concurrently with performing a visibility pass to generate visibility data for two or more bins of an image, a processing system determines whether a primitive to be rendered covers at least a predetermined threshold percentage of a tile of the image. In response to the primitive coving at least the predetermined threshold percentage of the tile, the processing system stores the depth data of the primitive in a depth buffer for pixel-based rendering. In response to the primitive not covering at least the predetermined threshold percentage of the tile, the processing system fuses the primitive with one or more preceding primitives sharing an edge with the primitive in the tile to generate a fused primitive. In response to the fused primitive being valid in the tile, the processing system passes the depth data of the fused primitive to the depth buffer.

Abstract: A system and method for improved graphics rendering in a video game environment.

Abstract: Image processing includes obtaining image I[0,0] of a picture captured by an image capture means, in a state where light is irradiated to the picture from a light source at a reference position relative to a normal line of the picture, obtaining image I[?1,0] of the picture captured by an image capture means, in a state where the light is irradiated to the picture from the light source at a position inclined from the reference position at an angle ?1 in the first direction, obtaining image I[0, ?1] of the picture captured by an image capture means, in a state where the light is irradiated to the picture from the light source at a position inclined by an angle ?1 from the reference position in a second direction different from the first direction, creating a three-dimensional map of the picture, using a set of images I[0, ?1] and I[0, ?2], merging at least a part of each of image I[?1,0], image I[0,?1], and image I[0,?2] with respect to image I[0,0], and recording as two-dimensional image data the image subjec

Abstract: Graphics processing system configured to perform ray tracing. Rays are bundled together and processed together. When differential data is needed by a shader, the data of a true ray in the bundle can be used rather than processing separate tracker rays.

Abstract: A method and device with light estimation are provided. A method performed by an electronic device includes generating a reference image based on image data acquired by capturing a reference object and based on a first image signal processing (ISP) setting, generating a background image based on raw image data acquired by capturing a real background in which the reference object is positioned and based on a second ISP setting, estimating light information corresponding to the background image using a light estimation model, rendering a virtual object image corresponding to the light information and the reference object, and training the light estimation model based on a difference between the reference image and the virtual object image.

Abstract: Disclosed is a display apparatus and a method for digital manipulation of a virtual user interface in virtual environments. The display apparatus includes a light source, tracking means, and a processor configured to control the light source to project the virtual user interface, and process tracking data from the tracking means to detect the proximity of an interaction element, such as a user's hand, to an invisible segment of a virtual widget within the virtual user interface. Upon proximity detection, the light source is activated to display the virtual widget. The processor then determines if the segment has been activated and subsequently processes any positional changes of the interaction element to adjust the virtual user interface accordingly. This adjustment is performed in accordance with predefined visual effects that correspond to the activated segment, enabling intuitive and dynamic user interaction.

Abstract: Ray tracing units, processing modules and methods are described for generating one or more reduced acceleration structures to be used for intersection testing in a ray tracing system for processing a 3D scene. Nodes of the reduced acceleration structure(s) are determined, wherein a reduced acceleration structure represents a subset of the 3D scene. The reduced acceleration structure(s) are stored for use in intersection testing. Since the reduced acceleration structures represent a subset of the scene (rather than the whole scene) the memory usage for storing the acceleration structure is reduced, and the latency in the traversal of the acceleration structure is reduced.

Abstract: Graphics processing systems can include lighting effects when rendering images. “Light probes” are directional representations of lighting at particular probe positions in the space of a scene which is being rendered. Light probes can be determined iteratively, which can allow them to be determined dynamically, in real-time over a sequence of frames. Once the light probes have been determined for a frame then the lighting at a pixel can be determined based on the lighting at the nearby light probe positions. Pixels can then be shaded based on the lighting determined for the pixel positions.

Abstract: A ray tracing unit and method for processing a ray in a ray tracing system performs intersection testing for the ray by performing one or more intersection testing iterations. Each intersection testing iteration includes: (i) traversing an acceleration structure to identify the nearest intersection of the ray with a primitive that has not been identified as the nearest intersection in any previous intersection testing iterations for the ray; and (ii) if, based on a characteristic of the primitive, a traverse shader is to be executed in respect of the identified intersection: executing the traverse shader in respect of the identified intersection; and if the execution of the traverse shader determines that the ray does not intersect the primitive at the identified intersection, causing another intersection testing iteration to be performed. When the intersection testing for the ray is complete, an output shader is executed to process a result of the intersection testing for the ray.

Abstract: A system for computational estimation sampling from non-trivial probability distributions. The system comprises a processor, operating in conjunction with computer memory. The processor is configured to conduct importance sampling using normalizing flows where a base distribution has a set of parameters that can be adjusted to account for heavy-tailed distributions.

Abstract: A generation apparatus includes following units. A first acquisition unit acquires a first virtual viewpoint image generated based on a plurality of images captured by a plurality of image capturing apparatuses, and a first virtual viewpoint in a three-dimensional space including an image capturing space captured by the plurality of image capturing apparatuses. A second acquisition unit acquires a second virtual viewpoint image generated based on the plurality of captured images, and a second virtual viewpoint arranged with respect to a predetermined surface in the three-dimensional space and the first virtual viewpoint. A generation unit generates a third virtual viewpoint image corresponding to the first virtual viewpoint and including an image indicating a state where an object in the image capturing space is reflected on the predetermined surface, based on the first virtual viewpoint image, and the second virtual viewpoint image.

Abstract: An image rendering method and a related apparatus, terminal and storage medium that: obtains a low-resolution first shadow map and a high-resolution second shadow map of a target virtual object, determines a first texel range of each texel and a central point in the first shadow map, performs scaling processing, based on the central point, on the first texel range by using a scaling coefficient to obtain a second texel range, and performs shadow rendering on the target virtual object according to the second texel range. A shadow map-based image rendering process is implemented, and since a calculation amount of determining a map position according to a low-resolution map is relatively small, image rendering efficiency is improved. Additionally, rendering of a high-resolution shadow map is completed, generation of edge aliasing is avoided, and an image rendering effect is improved.

Abstract: The invention relates to a computer implemented method for rendering a 2D/3D model. The method comprises the step of providing a memory unit and a first effect unit, the memory unit configured to store data regarding the model and send the data into at least the first effect unit, the first effect unit configured to receive the sent data, render the model based on the received data and generate a first rendering result; generating a second effect unit for performing a rendering process; arranging the second effect unit such that the second effect unit is configured to receive at least one of the first rendering result and the data from the memory unit; detecting a change of the data stored in the memory unit; and rendering, by the first and second effect units, the 2D/3D model.

Abstract: Example embodiments described herein relate to an augmented-reality system to generate and cause display of interactive augmented reality content at a client device.

Abstract: Aspects of this disclosure relate to a process for rendering graphics that includes performing, with a hardware unit of a graphics processing unit (GPU) designated for vertex shading, a vertex shading operation to shade input vertices so as to output vertex shaded vertices, wherein the hardware unit adheres to an interface that receives a single vertex as an input and generates a single vertex as an output. The process also includes performing, with the hardware unit of the GPU designated for vertex shading, a hull shading operation to generate one or more control points based on one or more of the vertex shaded vertices, wherein the one or more hull shading operations operate on at least one of the one or more vertex shaded vertices to output the one or more control points.

Abstract: Reducing data used during capture in a physical capture volume by selectively activating image capture devices from a virtual view, including: setting up a virtual camera to receive information about and visualize the physical capture volume and a plurality of image capture devices in the virtual view; providing, to the virtual camera, the virtual view of the physical capture volume with a capability to move around the physical capture volume and activate or deactivate each of the plurality of image capture devices; calculating a view frustum, wherein the view frustum is a region of 3-D space within the physical capture volume that would appear on a view screen of the virtual camera; and defining the view frustum of the virtual camera which intersects with the plurality of image capture devices defined in the virtual view.

Abstract: Systems and methods for detecting roadway and/or pathway lighting conditions are disclosed. According to certain aspects, various devices or sensors may collect a set of lighting measurements respectively at a set of locations along a roadway and/or pathway, and capture scanning data indicative of a set of surroundings located along the roadway and/or pathway. Further, an electronic device may analyze the scanning data to classify a set of light poles along the roadway and/or pathway as well as associate the set of lighting measurements with the set of locations. Moreover, the electronic device may generate an electronic file indicating the set of lighting measurements associated with the set of locations and the set of light poles classified at the portion of the set of locations.

Abstract: A wearable device may include a head-mounted display (HMD) for rendering a three-dimensional (3D) virtual object which appears to be located in an ambient environment of a user of the display. The relative positions of the HMD and one or more eyes of the user may not be in desired positions to receive, or register, image information outputted by the HMD. For example, the HMD-to-eye alignment may vary for different users and may change over time (e.g., as a user moves around and/or the HMD slips or is otherwise displaced). The wearable device may determine a relative position or alignment between the HMD and the user's eyes. Based on the relative positions, the wearable device may determine if it is properly fitted to the user, may provide feedback on the quality of the fit to the user, and may take actions to reduce or minimize effects of any misalignment.

Abstract: A method is performed at an electronic device including one or more processors and a non-transitory memory. The method includes obtaining a first spherical Gaussian (SG) lobe that characterizes ambient light from a physical environment. The method may include determining the first SG lobe based on a 360 degree image of the physical environment. The first SG lobe indicates a first directional characteristic associated with the ambient light. The method includes determining a first plurality of sampling rays based on the first directional characteristic. The method includes obtaining a depth value that is associated with a computer-generated object. The depth value may be from a depth buffer, which is populated with the depth value during rendering of the computer-generated object. The method includes generating a shadow that is associated with the computer-generated object, based on the depth value and a first sampling ray of the first plurality of sampling rays.

Abstract: A method of rendering an image of a 3-D scene includes rendering a noisy image at a first resolution; obtaining one or more guide channels at the first resolution, and obtaining one or more corresponding guide channels at a second resolution. The second resolution may be the same resolution as, or a higher resolution than, the first resolution. For each of a plurality of local neighbourhoods, the method comprises: calculating the parameters of a model that approximates the noisy image as a function of the one or more guide channels (at the first resolution), and applying the calculated parameters to the one or more guide channels at the second resolution, to produce a denoised image at the second resolution.

Abstract: In various examples, a virtual light meter may be implemented along with ray tracing techniques in order to determine incident light values—e.g., incoming irradiance, incident radiance, etc.—for adjusting auto exposure values of rendered frames. For example, one or more rays may be used to sample incident light over a sampling pattern—such as a hemispherical sampling pattern—for any position in a virtual game environment. As a result, the incident light values may be sampled near a subject of interest in a scene or frame such that exposure values are consistent or stable regardless of the composition of the rendered frames.

Abstract: A learning system (10) comprises: a generation unit (110) configured to synthesize with an original image including a living body, information on a reflection component of light not derived from the original image to generate a synthetic image; and a learning unit (130) configured to perform learning for an estimation model (121) based on information on a specific portion of the living body estimated from the synthetic image by the estimation model and correct answer information showing a correct answer of the information on the specific portion in the original image. According to this learning system, it is possible to learn the estimation model appropriately.

Abstract: A method of dynamic media generation for presentation via projection on an animated figure includes defining, via processing circuitry, a computer-generated model of the animated figure, operating, via the processing circuitry, a manufacturing system to generate a tangible model based on the computer-generated model, generating, via the processing circuitry, a revised computer-generated model based on the tangible model, simulating, via the processing circuitry, projection of imagery onto the revised computer-generated model, and operating, via the processing circuitry, a projector to project the imagery onto the tangible model based on simulated projection of the imagery onto the revised computer-generated model.

Abstract: Systems for transitioning a user interface arrangement from a display of a two-dimensional image of a user to a rendering of a three-dimensional representation of the user is provided. A system can start with a UI including a rendering of a user that is based on a 2D image file. The system can receive an input that is configured to cause the system to transition the display of the rendering of the 2D image of the select user to a rendering of the three-dimensional representation of the select user. To display the rendering of the 3D representation of the select user, the system uses permission data and a three-dimensional model defining a position and orientation to display the 3D representation of the user. The system allows users to switch between viewing modes to allow users to interact with content using the most effective type of hardware.

Abstract: One or more rotated bounding volumes are generated for one or more nodes of a bounding volume hierarchy (BVH). Volume intersection ray tracing tests are then be performed using the rotated bounding volumes with the aim of reducing the number of calculations required relative to an original, non-rotated bounding volume. Rotated bounding volumes are selected from a plurality of candidate rotations, and selection of one of the candidate rotations are based on surface areas, such as minimum total surface areas, of bounding volumes corresponding to each of the candidate rotations. In order to minimize data storage and increase performance, a number of candidate rotations may be limited to a predetermined set of rotations.

Abstract: Systems and methods of the present disclosure enable automated roof planning using a processor. The processor receives a digital image of a roof of a structure and models each roof plane of the roof to generate a roof model. The processor determines dimensions of each roof plane based on the roof model. The processor retrieves roofing accessory data from a database, the roofing accessory data solar roofing accessory part identifiers and solar roofing accessory part performance characteristics for solar roofing accessories. The processor simulates multiple candidate roof layouts based on the dimensions of each roof plan and the solar roofing accessory parts and determines a utilization prediction for each candidate layout. Based on each utilization prediction, the processor determines a particular roof layout having selected solar roofing accessory parts, and generates a solar roof design, including a list of materials, for the particular roof layout.

Abstract: The present disclosure relates to systems, methods, and non-transitory computer-readable media that generate a height map for a digital object portrayed in a digital image and further utilizes the height map to generate a shadow for the digital object. Indeed, in one or more embodiments, the disclosed systems generate (e.g., utilizing a neural network) a height map that indicates the pixels heights for pixels of a digital object portrayed in a digital image. The disclosed systems utilize the pixel heights, along with lighting information for the digital image, to determine how the pixels of the digital image project to create a shadow for the digital object. Further, in some implementations, the disclosed systems utilize the determined shadow projections to generate (e.g., utilizing another neural network) a soft shadow for the digital object. Accordingly, in some cases, the disclosed systems modify the digital image to include the shadow.

Abstract: Disclosed herein are various embodiments related generally to computer vision, graphics, image scanning, shadow casters, shadow caster scanners, and image processing as well as associated mechanical, electrical and electronic hardware, computer software and systems, and wired and wireless network communications to form at least three-dimensional models or images of objects and environments.

Abstract: An angular direction identifying device includes: a photographed image acquiring unit configured to acquire a photographed image by a camera provided in a movable body; an angular direction information acquiring unit configured to acquire angular direction information in a case where a predetermined physical object suggesting an angular direction is shown in the photographed image, the angular direction information indicating the angular direction suggested by the predetermined physical object; and a view-axis angular direction identifying unit configured to identify an angular direction in which a view axis of the camera is pointed, based on the angular direction information and a position relation between the predetermined physical object in the photographed image and the view axis of the camera.

Abstract: A medical image processing apparatus comprises processing circuitry configured to: receive a selection of a preset transfer function that sets colors used in volume rendering, wherein the preset transfer function is selected for use in rendering at least one target material; determine a first transfer function relating to transmission or extinction color, wherein the determining of the first transfer function is based on the preset transfer function; generate first rendering data based on transmission or extinction color by using the first transfer function; and determine a second transfer function relating to reflection color, wherein the determining of the second transfer function is based on the first rendering data, wherein the first transfer function and second transfer function are for use in global illumination rendering.

Abstract: An alternate root tree or graph structure for ray and path tracing enables dynamic instancing build time decisions to split any number of geometry acceleration structures in a manner that is developer transparent, nearly memory storage neutral, and traversal efficient. The resulting traversals only need to partially traverse the acceleration structure, which improves efficiency. One example use reduces the number of false positive instance acceleration structure to geometry acceleration structure transitions for many spatially separated instances of the same geometry.

Abstract: Devices and methods for node traversal for ray tracing are provided, which comprise casting a first ray in a space comprising objects represented by geometric shapes, traversing, for the first ray, at least one first node of an accelerated hierarchy structure representing an approximate volume of a group of the geometric shapes and a second node representing a volume of one of the geometric shapes, casting a second ray in the space, selecting, for the second ray, a starting node of traversal based on locations of intersection of the first ray and the second ray and an identifier which identifies one or more nodes intersected by the first ray and traversing, for the second ray, the accelerated hierarchy structure beginning at the starting node of traversal.

Abstract: Methods and systems are disclosed for performing operations for applying augmented reality elements to a person depicted in an image. The operations include receiving an image that includes data representing a depiction of a person; generating a segmentation of the data representing the person depicted in the image; extracting a portion of the image corresponding to the segmentation of the data representing the person depicted in the image; applying a machine learning model to the portion of the image to predict a surface normal tensor for the data representing the depiction of the person, the surface normal tensor representing surface normals of each pixel within the portion of the image; and applying one or more augmented reality (AR) elements to the image based on the surface normal tensor.

Abstract: In an embodiment, a method (100) is described. The method comprises obtaining (102) a three-dimensional representation of a body surface. The method further comprises obtaining illumination information for the three-dimensional representation that is indicative of an orientation of the body surface relative to a reference axis. The method further comprises determining illumination compensation information (104). The illumination compensation information is used (106) to compensate for an illumination variation apparent from the illumination information in an image of the body surface.

Abstract: Various implementations disclosed herein include devices, systems, and methods that provide a CGR environment in which virtual objects from one or more apps are included. User interactions with the virtual objects are detected and interpreted by a system that is separate from the apps that provide the virtual objects. The system detects user interactions received via one or more input modalities and interprets those user interactions as events. These events provide a higher-level, input modality-independent, abstractions of the lower-level input-modality dependent user interactions that are detected. The system uses UI capability data provided by the apps to interpret user interactions with respect to the virtual object provided by the apps. For example, the UI capability data can identify whether a virtual object is moveable, actionable, hover-able, etc. and the system interprets user interactions at or near the virtual object accordingly.

Abstract: Virtual creation and visualization of a physically based virtual material swatch includes identifying a capture device model to support for use in the generation of the virtual material swatch, determining camera response of the capture device model, determining lookup tables of linearly transformed cosines for a plurality of parameters of a shading model, identifying a sample physical material, and determining material description for the sample physical material. A system and method may also include scanning a user physical material, selecting one or more of material preset, geometry preset, environment preset, and camera preset, and creating a virtual material swatch for the user physical material based on the selections.

Abstract: A three-dimensional (3D) sensing device is configured to sense an object. The 3D sensing device includes a flood light source, a structured light source, an image sensor, and a controller. The controller is configured to perform: commanding the flood light source and the structured light source to emit a flood light and a structured light in sequence; commanding the image sensor to sense a first reflective light and a second reflective light in sequence, so as to obtain a first image frame and a second image frame; combining the first image frame and the second image frame into a determination frame; and determining that the object is a specular reflection object in response to determining that the determination frame has at least two spots having gray levels satisfying a predetermined condition. A specular reflection object detection method is also provided.

Abstract: According to an aspect of an embodiment, operations may comprise receiving sensor data from one or more vehicles, determining, by combining the received sensor data, a high definition map comprising a point cloud, and labeling one or more objects in the point cloud. The operations may also comprise generating training data by receiving a new image captured by one of the vehicles, receiving a pose of the vehicle when the new image was captured, determining an object having a label in the point cloud that is observable from the pose of the vehicle, determining a position of the object in the new image, and labeling the new image by assigning the label of the object to the new image, the labeled new image comprising the training data. The operations may also comprise training a deep learning model using the training data.