Nvidia Patent | Motion Adaptive Rendering Using Variable Rate Shading

Patent: Motion Adaptive Rendering Using Variable Rate Shading

Publication Number: 20200051290

Publication Date: 20200213

Applicants: Nvidia

Abstract

Motion adaptive shading increases rendering performance for real-time animation in graphics systems while maintaining dynamic image quality. Each frame of an animation is statically displayed within a refresh interval, while a viewer’s eyes move continuously relative to the image when actively tracking a moving object being displayed. As a result, a statically displayed frame is essentially smeared across the viewer’s continuously moving retina over the lifetime of the frame, causing a perception of blur referred to as an eye-tracking motion blur effect. A region of an image depicting a moving object may be rendered at a lower shading rate because eye-tracking motion blur will substantially mask any blur introduced by reducing the shading rate. Reducing an average shading rate for rendering frames reduces computational effort per frame and may advantageously allow a rendering system to operate at a higher frame rate to provide a smoother, clearer visual experience.

CLAIM OF PRIORITY

[0001] This application claims the benefit of U.S. Provisional Application No. 62/716,497 (Attorney Docket No. 18-SC-0232US01) titled “Motion Adaptive Rendering Using Variable Rate Shading,” filed Aug. 9, 2018, the entire contents of which is incorporated herein by reference.

TECHNICAL FIELD

[0002] The present disclosure relates to real-time graphics rendering, and more particularly to motion adaptive rendering using variable rate shading.

BACKGROUND

[0003] Liquid crystal display (LCD) systems and organic light-emitting diode (OLED) display systems are known to introduce an undesirable eye-tracking motion blur effect when displaying motion or animation. The effect is caused by each frame being statically displayed within a refresh interval, whereas a viewer’s eyes move continuously relative to the image when actively tracking a moving object being displayed. As a result, a statically displayed frame is essentially smeared across the viewer’s continuously moving retina over the lifetime of the frame, causing a perception of blur. The eye-tracking motion blur effect can significantly reduce visual quality at common interactive frame rates of 60 fps, and can be noticeable at 120 fps.

[0004] Interactive graphics systems provide finite real-time rendering capacity, forcing a trade-off between static image quality and frame rate. This trade-off space conventionally reduces static image quality at higher frame rates or increases motion blur at lower frame rates. There is a need for addressing these issues and/or other issues associated with the prior art.

SUMMARY

[0005] Motion adaptive rendering (or motion adaptive shading) increases rendering performance for real-time animation in graphics systems while maintaining dynamic image quality. Each frame of an animation is statically displayed within a refresh interval, while a viewer’s eyes move continuously relative to the image when actively tracking a moving object being displayed. As a result, a statically displayed frame is essentially smeared across the viewer’s continuously moving retina over the lifetime of the frame, causing a perception of blur referred to as an eye-tracking motion blur effect. A region of an image depicting a moving object may be rendered at a lower shading rate because eye-tracking motion blur will substantially mask any blur introduced by reducing the shading rate. Reducing an average shading rate for rendering frames reduces computational effort per frame and may advantageously allow a rendering system to operate at a higher frame rate to provide a smoother, clearer visual experience. Reducing the average shading rate may also allow a rendering system that is not frame rate limited to operate at lower power. Motion adaptive shading applies generally to rendering graphics primitives, including two-dimensional (2D) graphics primitive and three-dimensional (3D) graphics primitives.

[0006] A method, computer readable medium, and system are disclosed for motion adaptive rendering, comprising receiving motion data for a region of a frame, calculating a per-region motion value from the motion data, determining a shading rate for the region based on the per-region motion value, and rendering a graphics primitive at the shading rate to produce color data for the region.

[0007] A method, computer readable medium, and system are disclosed for content-and-motion adaptive rendering, comprising receiving motion data for a region of a frame, calculating a per-region motion value from the motion data, determining a shading rate for the region based on the per-region motion value and pixel color data from a previously rendered frame, and rendering a graphics primitive at the shading rate to produce color data for the region.

BRIEF DESCRIPTION OF THE DRAWINGS

[0008] FIG. 1A illustrates a flowchart of a method for motion adaptive rendering, in accordance with an embodiment.

[0009] FIG. 1B illustrates different pixel block sizes for different tiles, in accordance with an embodiment.

[0010] FIG. 1C illustrates a rendered frame depicting animated objects, in accordance with an embodiment.

[0011] FIG. 1D illustrates a rendered frame depicting camera motion, in accordance with an embodiment.

[0012] FIG. 1E illustrates a rendered static image from a graphics scene, in accordance with an embodiment.

[0013] FIG. 1F illustrates a rendered image of a graphics scene with simulated eye-tracking motion blur at 60 frames per second refresh rate, in accordance with an embodiment.

[0014] FIG. 1G illustrates a rendered image of a graphics scene with simulated eye-tracking motion blur at 120 frames per second refresh rate using motion adaptive shading to reduce shading effort, in accordance with an embodiment.

[0015] FIG. 1H illustrates a comparison between motion adaptive shading and conventional rendering using full shading effort, in accordance with an embodiment.

[0016] FIG. 2 illustrates a flowchart of a method for determining a shading rate by jointly considering content variation and motion, in accordance with an embodiment.

[0017] FIG. 3 illustrates a parallel processing unit, in accordance with an embodiment.

[0018] FIG. 4A illustrates a general processing cluster within the parallel processing unit of FIG. 3, in accordance with an embodiment.

[0019] FIG. 4B illustrates a memory partition unit of the parallel processing unit of FIG. 3, in accordance with an embodiment.

[0020] FIG. 5A illustrates the streaming multi-processor of FIG. 4A, in accordance with an embodiment.

[0021] FIG. 5B is a conceptual diagram of a processing system implemented using the PPU of FIG. 3, in accordance with an embodiment.

[0022] FIG. 5C illustrates an exemplary system in which the various architecture and/or functionality of the various previous embodiments may be implemented.

[0023] FIG. 6 is a conceptual diagram of a graphics processing pipeline implemented by the PPU of FIG. 3, in accordance with an embodiment.

DETAILED DESCRIPTION

[0024] Techniques disclosed herein, collectively referred to as motion adaptive shading, increase rendering performance for real-time animation in 2D and 3D graphics systems while maintaining dynamic image quality. A given rendered frame may be organized into tiles, and each tile may be rendered according to tile-specific shading rates in one or more dimensions. A given rendered frame may also be composed of geometry primitives (e.g. triangles), and each primitive may be rendered according to a primitive-specific shading rate in one or more dimensions. Tiles and primitives are hereby collectively referred to as regions. A shading rate for a region may be calculated based on motion data (e.g., motion vectors) within the region (for motion adaptive shading), or on content variation such as luminance and/or color frequency and/or contrast within the region (for content adaptive shading). Furthermore, the shading rate may be calculated using both motion data and content variation. A shading rate refers to a number of shaded color samples per pixel. In an embodiment, for example, a shading rate of two indicates that pixels are oversampled to include two shaded color samples each, while a shading rate of one half indicates that one shaded color sample is applied to two pixels. Shading rate may be directional, such as along X and Y dimensions.

[0025] Motion adaptive shading allows rendering systems to overcome conventional trade-off constraints between rendered static image quality and rendered frame rate by selectively reducing shading rates for regions that depict sufficient motion. A given region depicting pixel motion above one or more velocity thresholds may be rendered with correspondingly reduced shading rates because eye-tracking motion blur will substantially mask blur introduced by the reduced shading rates within the region. Reducing an average shading rate for an animated sequence reduces computational effort per frame and may advantageously allow for a higher frame rate. In an embodiment, geometry sampling frequency and/or precision of primitive edges are determined independently from shading rates and are therefore not affected by shading rate.

[0026] To reduce potential loss of visual quality, shading rates for each region are individually calculated such that the entailed reduction of visual detail is primarily masked by perceived blur. In practice, motion adaptive shading produces visual quality similar to ground truth observation. Perceived blur at each region is caused by eye-tracking motion blur, any motion blur effect computed in a rendering engine, or a combination thereof. When a given frame is rendered, shading rates are adaptively calculated for each region prior to or during a color shading pass for the frame.

[0027] FIG. 1A illustrates a flowchart of a method 110 for motion adaptive rendering (shading), in accordance with an embodiment. Although method 110 is described in the context of a processing unit, the method 110 may also be performed by a program, custom circuitry, or by a combination of custom circuitry and a program. For example, the method 110 may be executed by a GPU (graphics processing unit), CPU (central processing unit), or any processor capable of performing variable rate shading. Furthermore, persons of ordinary skill in the art will understand that any system that performs method 110 is within the scope and spirit of embodiments of the present disclosure.

[0028] At step 111, the processing unit receives motion data for a region of a frame. The frame may be one in a sequence of animation frames. In some embodiments, the region may comprise a tile having a fixed or variable size within the frame. In other embodiments, the region may comprise the covered pixels of a geometry primitive, such as a triangle. The motion data may comprise motion vectors for the region. For example, the motion data may include a motion vector for each pixel within the region. In an embodiment, the motion data is calculated according to a camera position change or object position change between an immediately prior frame and the frame. The motion data may be computed using any technically feasible technique(s). Furthermore, the motion data may be computed by the processing unit.

[0029] At step 113, the processing unit calculates a per-region motion value from the motion data. In an embodiment, the per-region motion value includes at least one pixel velocity represented in units of pixels (e.g., normalized distance) per frame (e.g., normalized frame time). In an embodiment, the pixel velocity and/or per-region motion value may be calculated assuming frame time is constant, although frame time may actually vary depending on rendering effort. In other embodiments, pixel velocity and/or per-region motion value may be calculated to account for a rendered frame time. In an embodiment, the motion data comprises an individual motion vector for each pixel within the region, and a given per-region motion value is calculated based on the individual motion vectors. In another embodiment, the motion data comprises a motion vector for each vertex of a geometry primitive within the region.

[0030] In certain embodiments, calculating a per-region motion value from the motion data comprises selecting a minimum motion vector value from the individual motion vectors. In other embodiments, different techniques may be implemented to calculate a per-region motion value for the region. For example, in an embodiment, a per-region motion value may be calculated to be an average, median, or mode value from the individual motion vectors. In another embodiment, a per-region motion value may be calculated as a specified percentile velocity of the individual motion vectors.

[0031] In an embodiment, independent per-region motion values are calculated per display dimension. In such an embodiment, a per-region motion value is calculated along an X (horizontal) dimension and, separately a per-region motion value is calculated along a Y (vertical) dimension. For example, a per-region motion value for the X dimension may be calculated as a minimum component value from all X components of the individual motion vectors; and a per-region motion value for the Y dimension may be calculated as a minimum value from all Y components of the individual motion vectors.

[0032] At step 115, the processing unit determines a shading rate for the region based on the per-region motion value. Any technically feasible technique may be performed to determine the shading rate. In an embodiment, the shading rate specifies a value of one (i.e. one shaded sample per pixel), one half (i.e., one shaded sample per two pixels), or one quarter (i.e., one shaded sample per four pixels). In other embodiments, different or additional shading rates may be specified. Furthermore, a shading rate may be calculated for each dimension.

[0033] In an embodiment, a first velocity threshold (V2) and a second velocity threshold (V4) are defined where V2 is less than V4. When a per-region motion value along the X dimension is calculated to be equal to or below the first velocity threshold, the shading rate along the X dimension is determined to be one. When the per-region motion value is above the first velocity threshold but equal to or below the second velocity threshold, the shading rate along the X dimension is determined to be one half (i.e. half-rate shading). When the per-region motion value is above the second velocity threshold, the shading rate along the X dimension is determined to be one quarter (i.e. quarter-rate shading). Furthermore, shading rates along the Y dimension may be similarly calculated.

[0034] In an embodiment, each region comprises a tile, and each tile may be further organized into pixel blocks. An exemplary tile may comprise a 16.times.16 set of pixels having a pixel block size of 1.times.1, 1.times.2, 2.times.1, 1.times.4, 4.times.1, 2.times.4, 4.times.2, or 4.times.4 pixels. Pixel block size may be determined based on shading rates for the tile. Shading rate along each dimension determines a pixel block size. For example, if a particular tile is determined to have a shading rate of one half in the vertical dimension and one quarter in the horizontal dimension, then a 4.times.2 pixel block size may be used.

[0035] In an embodiment, a region is covered at least partially by a geometry primitive such as a triangle. The processing unit may rasterize the primitive into individual pixels according to geometric coverage of the primitive, but shading of the individual pixels may proceed according to pixel block sizes of 1.times.1, 1.times.2, 2.times.1, 1.times.4, 4.times.1, 2.times.4, 4.times.2, or 4.times.4 pixels. In such an embodiment, each pixel in a particular pixel block is assigned the same shading result.

[0036] At step 117, the processing unit renders 3D graphics primitives at the shading rate to produce color data for the region. Any technically feasible technique may be performed to render the 3D graphics primitives, including pixel shading in a forward or deferred rendering pipeline, screen-space rendering using compute shaders, ray tracing, and/or decoupled shading. A shading operation may include, without limitation, generating and combining one or more samples, such as texture mapping samples, sub-pixel samples, or ray trace samples.

[0037] In an embodiment, method 110 is performed on a first tile to determine a first shading rate(s) for the first tile, and method 110 is subsequently performed on a second, adjacent tile to determine a second shading rate(s) for the second tile. In certain embodiments, the first shading rate and the second shading rate are limited to a maximum shading rate difference. In an embodiment, the maximum shading rate difference may be specified as one list position difference from a list of possible shading rates (e.g., one, one half, one quarter, and so forth). For example, if the first shading rate is determined to be one, then the second shading rate may be either one or one half, but not one quarter (list position difference of two).

[0038] In certain embodiments, the processing unit includes hardware subsystems configured to perform variable rate shading at one or more positions within a rendering pipeline. In an embodiment, the processing unit is configured to execute shaders that perform variable rate shading, with a shading rate specified individually for each tile or each primitive.

[0039] More illustrative information will now be set forth regarding various optional architectures and features with which the foregoing framework may be implemented, per the desires of the user. It should be strongly noted that the following information is set forth for illustrative purposes and should not be construed as limiting in any manner. Any of the following features may be optionally incorporated with or without the exclusion of other features described.

[0040] FIG. 1B illustrates different pixel block sizes for different tiles 120, 122, 124, in accordance with an embodiment. As shown, tile 120 has a pixel block size of one pixel, tile 122 has a pixel block size of 1.times.4 pixels, and tile 124 has a pixel block size of 2.times.2 pixels. Additional pixel block sizes may also be specified, potentially up to the size of the tile. In an embodiment, pixel block sizes include 1.times.1, 1.times.2, 1.times.4, 2.times.1, 2.times.2, 2.times.4, 4.times.1, 4.times.2, and 4.times.4 within a 16.times.16 tile. Of course, different tile sizes and different pixel block sizes may be implemented instead without departing the scope and spirit of various embodiments.

[0041] In an embodiment, a pixel block size of 1.times.1 includes one pixel and is shaded from at least one color shading operation. A pixel block of 4.times.4 pixels comprises sixteen pixels, which are collectively shaded from one color shading operation, thereby reducing computational workload relative to a 1.times.1 pixel block. Furthermore, a 16.times.16 tile comprising 4.times.4 pixel blocks only requires sixteen shading operations rather than two-hundred fifty-six color shading operations.

[0042] FIG. 1C illustrates a rendered frame 130 depicting animated objects 135, in accordance with an embodiment. As shown, the rendered frame 130 comprises a grid of adjacent square tiles organized along X and Y dimensions. For example, a Tile(0,0) is located at X=0, Y=0, and so forth. Each animated object 135(1), 135(2), 135(3) is shown moving at a different velocity and direction, with sequential frame-to-frame movement depicted as trailing dashed impressions, and velocity depicted as an associated arrow having direction and magnitude. As shown, animated object 135(1) is moving slowly along the Y (vertical) dimension, animated object 135(2) is moving quickly along the X (horizontal) dimension, and animated object 135(3) is moving diagonally along the X and Y dimensions. Each different animated object 135 may be perceived as having varying degree of eye-tracking motion blur, with faster-moving objects having more eye-tracking motion blur.

[0043] In an exemplary real-time frame sequence, velocity threshold V2 is set at three pixels per frame and velocity threshold V4 is set at six pixels per frame. Animated object 135(1) may be traveling at a screen space velocity of one pixel per frame along the Y dimension (less than V2) and zero pixels per frame along the X dimension (less than V2); consequently a shading rate for associated tiles (e.g. covered tiles) would be determined to be one in both X and Y dimensions, and shading would be performed on 1.times.1 pixel blocks within the associated tiles. Additionally, animated object 135(2) may be traveling at a screen space velocity of seven pixels per frame along the X dimension (greater than V4) and zero pixels per frame along the Y dimension (less than V2); consequently a shading rate for associated tiles would be determined to be one quarter in the X dimension and one in the Y dimension, and shading would be performed on 1.times.4 pixel blocks within the associated tiles. Finally, animated object 135(3) may be traveling at four pixels per frame in both X and Y dimensions (between V2 and V4); consequently a shading rate for associated tiles would be determined to be one half in each dimension and shading would be performed on 2.times.2 pixel blocks within the associated tiles.

[0044] In an embodiment, a processing unit is configured to generate the rendered frame 130 by performing techniques disclosed herein, including method 110 of FIG. 1A, to reduce overall shading computation associated with generating the rendered frame 130. More specifically, shading rates for each tile may be reduced according to motion within the tile, such that a shading rate reduction is substantially masked by inherent eye-tracking motion blur for the tile. In the present example, a tile at location (4,1) may be shaded at a reduced shading rate along the X dimension because animated object 135(2) covers all pixels in the tile and has a high pixel velocity. If the velocity of animated object 135(2) is above the first velocity threshold (e.g., V2 exceeds three pixels/frame), then pixels within the tile at location (4,1) may be shaded with a reduced shading rate of one half. If the velocity is above the second velocity threshold (e.g., V4 exceeds six pixels/frame), then a shading rate of one quarter may be used.

[0045] Similarly, a tile at location (1,5) may be shaded at reduced shading rates along both X and Y dimensions because animated object 135(3) covers all pixel in the tile and has moderate pixel velocity and both X and Y. Selectively reducing shading rates in this way creates a broader tradeoff space between static shaded resolution per frame and frame rate, allowing systems to achieve better overall visual quality. In an embodiment, pixels in a given tile are rendered according to a shading rate(s) for the tile. In another embodiment, pixels associated with a given primitive (e.g., comprising an object) are rendered according to a shading rate for the primitive.

[0046] FIG. 1D illustrates a rendered frame 130 depicting camera motion, in accordance with an embodiment. As shown, a camera is panning in a left to right motion, resulting in pixel motion throughout the frame 130, with different pixels within the frame 130 potentially subject to different pixel motion. In an embodiment, motion vectors for pixels within the frame 130 are computed based on differences between camera position at the current frame and a previous frame along with depth map values for the current frame. In an embodiment, a shader is configured to compute world-space positions for each pixel based on a corresponding depth value and a view-projection matrix. The world-space positions can be transformed based on a previous view-projection matrix of a previous frame to produce a viewport location in the previous frame of a scene geometry element that is now at a given pixel in the current frame. A velocity vector for a given pixel can be calculated as a difference in viewport position of the geometry between the current frame (at the pixel) and the previous frame. While this technique accounts for pixel velocity of stationary objects, pixel velocity values do not necessarily capture pixel velocity of objects moving in real-time through world-space.

[0047] Any region of the frame 130 may be subject to eye-tracking motion blur. When the camera is panning sufficiently quickly, most or all tiles within the frame 130 may be eligible for reduced shading rates, thereby allowing for higher frame rates and smoother, clearer perceived motion.

[0048] FIG. 1E illustrates a rendered static image from a graphics scene, in accordance with an embodiment. The static image represents the highest visual quality for an exemplary rendering platform.

[0049] FIG. 1F illustrates a rendered image of a graphics scene with simulated eye-tracking motion blur at 60 frames per second (FPS) refresh rate, in accordance with an embodiment. This image may be considered a baseline as 60 FPS displays are in common use in relevant real-time animation applications. Note that significant blurring may be seen along the X dimension due to camera panning along the X dimension.

[0050] FIG. 1G illustrates a rendered image of a graphics scene with simulated eye-tracking motion blur at 120 FPS refresh rate using motion adaptive shading to reduce shading effort, in accordance with an embodiment. As shown, a significantly less blurry image is produced at 120 FPS, a frame rate that may be achieved using motion adaptive shading to reduce color shading effort on the rendering platform. The rendering platform could not otherwise achieve 120 FPS.

[0051] FIG. 111 illustrates a comparison between motion adaptive shading and conventional rendering using full shading effort, in accordance with an embodiment. As shown, motion adaptive shading produces similar visual quality relative to shading performed with no shading rate reductions. However, the rendering platform is actually able to operate at 120 FPS using motion adaptive shading, whereas the same platform could not otherwise achieve 120 FPS using conventional techniques.

[0052] From the perspective of frequency domain signal analysis, eye-tracking motion blur is substantially equivalent to applying a first low-pass filter that is directional and oriented along the motion direction of a moving object (or frame region) being visually tracked. This first low-pass filter attenuates certain details in a visual signal having frequency content higher than a certain passing threshold (e.g., assuming an ideal low-pass filter). At the same time, reducing shading rate along the same direction also reduces high-frequency details in the signal (a second low pass filter applied to the same visual information). Therefore, when the passing threshold frequency associated with shading rate reduction (second low-pass filter) is higher than the passing threshold frequency associated with eye-tracking motion blur (first low-pass filter), the effect of shading rate reduction can be essentially masked (hidden). In practice, masking of a shading rate reduction is guaranteed by first determining the amount of eye-tracking motion blur for a given screen-space speed for rendered contents (e.g., animated objects 135, camera movement, etc.), and then applying a safe ratio of a shading rate reduction that can be masked by eye-tracking motion blur. In an embodiment, a first safe ratio is specified as a rate reduction of one half when pixel velocity is above approximately three pixels per frame; a second safe ratio is specified as a rate reduction of one quarter when pixel velocity is above approximately six pixels per frame.

[0053] Furthermore, slightly different velocity thresholds V2+ and V2- (or V4+ and V4-) may be directionally applied, such that when increasing a shading rate for a given tile in a subsequent frame, V2+ and V4+ may be used; and, when decreasing a shading rate for the tile in the subsequent frame, V2- and V4- may be used. Using directionally applied velocity thresholds provides hysteresis and may generally avoid unnecessarily rapid alternation between shading rates, which may be perceived by a viewer as undesirable visual artifacts. In this way, adding hysteresis may act as a damping function to avoid rapid oscillations or spikes in shading rate on the same tile or on a moving object.

[0054] In an embodiment, adjacent tiles are constrained to differ by one or less increments (one, one half, one quarter) of shading rate. In such an embodiment, a tile shading rate that is otherwise eligible to be reduced to be one quarter is only allowed to be reduced to one half when adjacent to a tile with a shading rate of one. In this way, a first shading rate for a first tile and a second shading rate for a second (adjacent) tile are limited to a maximum shading rate difference. Limiting shading rate differences for adjacent tiles serves to reduce or eliminate sharp changes in blur that can be perceived as undesirable visual artifacts.

[0055] The presently disclosed technique may be extended to support multiple shading operations per pixel, to effectively super sample a pixel shading result. In such an embodiment, additional velocity thresholds V1/2, V1/4, etc. may be defined to trigger shading rates of 2.times., 4.times., etc. samples per-pixel (super sampling). Super sampling may increase temporal stability and reduce aliasing artifacts appearing on high-frequency details. A higher per-pixel shading rate may provide additional detail and perceived image quality at lower object motion speeds, as aliasing artifacts are more easily noticeable at lower speeds. Such super sampled shading may be implemented in various shading pipelines.

发表评论

电子邮件地址不会被公开。 必填项已用*标注