Virtual digital human adaptive rendering method and device

Abstract

The disclosure provides a virtual digital person adaptive rendering method and device, and relates to the technical field of computers. The method comprises the following steps: obtaining initial weights of each region of a virtual digital person model and current interaction parameters, and updating the weight values according to the initial weights and the current interaction parameters; selecting initial LOD models for each region according to the weights and preset rules; further dynamically adjusting LOD precision levels of each region based on CPU loads and memory occupancies of the models; and performing differential rendering according to the adjusted LOD models. The disclosure realizes double adaptive allocation of rendering resources according to interaction attention and real-time loads, significantly improves the visual quality of a high-priority region under limited hardware resources, guarantees overall rendering fluency and stability, and is suitable for terminal environments with limited resources.

Description

Technical Field

[0001] This disclosure relates to the field of computer technology, and in particular to a method and apparatus for adaptive rendering of virtual digital humans. Background Technology

[0002] With the development of the metaverse and online interactive applications, the demand for rendering virtual digital humans on the web is growing. However, web browser environments typically suffer from limited memory and GPU computing power. Existing solutions, to ensure smooth operation, often employ a uniform low-quality rendering strategy for all parts of the virtual digital human, resulting in blurred facial details, a lack of realistic textures, and a poor user experience. In particular, it is difficult to achieve a reasonable allocation of rendering resources between high-attention areas (such as the face and hands) and low-attention areas, making it difficult to balance rendering quality and performance with limited resources. Therefore, there is an urgent need for a virtual digital human rendering method that can dynamically allocate rendering resources based on visual importance and achieve an adaptive balance between quality and web-based performance limitations. Summary of the Invention

[0003] This disclosure aims to at least partially address one of the technical problems in the related art.

[0004] Therefore, the first aspect of this disclosure proposes a virtual digital human adaptive rendering method, comprising:

[0005] Obtain the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model; The initial weights of each region are updated according to the interaction parameters to obtain the first weight of each region; Based on the first weight and the preset mapping rules, the initial LOD model corresponding to each region is selected from multiple LOD models of precision levels; The CPU load and memory usage of the virtual digital human model are determined based on the initial LOD model corresponding to each region. Based on the CPU load and / or memory usage, adjust the LOD model accuracy level corresponding to each region, and determine the LOD model corresponding to each region after adjustment; Based on the adjusted LOD models corresponding to each region, the corresponding regions of the virtual digital human model are rendered.

[0006] In some embodiments of this disclosure, the interaction parameters include at least one of the following: interaction state, distance between the user and the user, and display area ratio of each region.

[0007] In some embodiments of this disclosure, the interaction state includes a speaking state and a gesture state; updating the initial weights of each region according to the interaction state includes: in response to the virtual digital human model being in a speaking state, increasing the weight of the mouth region based on the initial weight of the mouth region in the virtual digital human model; in response to the virtual digital human model being in a gesture state, increasing the weight of the hand region based on the initial weight of the hand region in the virtual digital human model.

[0008] In some embodiments of this disclosure, updating the initial weights of each region based on the distance to the interacting user includes: reducing the facial region weight based on the initial weight of the facial region in the virtual digital human model in response to the distance to the interacting user being greater than or equal to a maximum distance threshold; and increasing the facial region weight based on the initial weight of the facial region in the virtual digital human model in response to the distance to the interacting user being less than or equal to a minimum distance threshold.

[0009] In some embodiments of this disclosure, updating the initial weights of each region based on the display area ratio of each region includes: determining a first target region whose display area ratio is less than or equal to a minimum ratio threshold among the plurality of regions; and reducing the weight of the first target region based on the initial weight of the first target region.

[0010] In some embodiments of this disclosure, the accuracy level and accuracy are positively correlated. Adjusting the accuracy level of the LOD model corresponding to each region based on the CPU load includes: in response to the CPU load being greater than or equal to the highest load threshold, reducing the accuracy level of the LOD model corresponding to each region based on the initial LOD model corresponding to each region; and in response to the CPU load being less than or equal to the lowest load threshold, increasing the accuracy level of the LOD model corresponding to each region based on the initial LOD model corresponding to each region.

[0011] In some embodiments of this disclosure, the accuracy level and accuracy are positively correlated. Adjusting the accuracy level of the LOD model corresponding to each region based on the memory usage includes: in response to the memory usage being greater than or equal to the highest memory threshold, determining a second target region among the plurality of regions whose first weight is less than a first weight threshold; and reducing the accuracy level of the LOD model corresponding to the second target region based on the initial LOD model corresponding to the second target region.

[0012] A second aspect of this disclosure provides a virtual digital human adaptive rendering apparatus, comprising: The acquisition module is used to acquire the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model. The weight update module is used to update the initial weight of each region according to the interaction parameters to obtain the first weight of each region; The first determining module is used to select the initial LOD model corresponding to each region from multiple LOD models of precision levels according to the first weight and the preset mapping rules. The second determining module is used to determine the CPU load and memory usage of the virtual digital human model based on the initial LOD model corresponding to each region; The adjustment module is used to adjust the accuracy level of the LOD model corresponding to each region based on the CPU load and / or the memory usage, and to determine the LOD model corresponding to each region after adjustment; The rendering module is used to render the corresponding areas of the virtual digital human model based on the adjusted LOD models corresponding to each area.

[0013] A third aspect of this disclosure provides an electronic device, including: a processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method described in the first aspect above.

[0014] A fourth aspect of this disclosure provides a computer-readable storage medium, characterized in that the computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method described in the first aspect above.

[0015] The adaptive rendering method for virtual digital humans disclosed herein dynamically updates the rendering weights of each region by introducing interaction parameters, and initially selects a Level of Detail (LOD) model based on these weights, achieving an initial match between rendering accuracy and visual attention. Based on the estimated CPU and memory load of the initially selected model, a secondary calibration of the LOD accuracy is performed, thus incorporating real-time system performance constraints into the decision-making loop. Under conditions of limited overall resources, it does not sacrifice model rendering quality but rather performs dual dynamic adjustments based on the current interaction scenario and actual hardware load, automatically and rationally allocating resources to prioritize the rendering quality of high-attention areas while suppressing resource consumption in non-critical areas. This effectively improves the overall visual expressiveness of the virtual digital human while maintaining stable and smooth system operation.

[0016] Additional aspects and advantages of this disclosure will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of this disclosure. Attached Figure Description

[0017] The above and / or additional aspects and advantages of this disclosure will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, in which: Figure 1 A flowchart illustrating an adaptive rendering method for virtual digital humans provided in this embodiment of the disclosure; Figure 2 This is a schematic diagram of a virtual digital human adaptive rendering device provided in an embodiment of the present disclosure. Detailed Implementation

[0018] Embodiments of this disclosure are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this disclosure, and should not be construed as limiting this disclosure.

[0019] Specifically, the virtual digital human adaptive rendering method and apparatus of the present disclosure are described below with reference to the accompanying drawings.

[0020] Figure 1 This is a flowchart illustrating an adaptive rendering method for virtual digital humans provided in an embodiment of this disclosure. Figure 1 As shown, the virtual digital human adaptive rendering method may include the following steps: Step 101: Obtain the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model.

[0021] In some embodiments of this disclosure, the virtual digital human model can be region-divided based on its skeletal topology and semantic functions. Specifically, according to the skeletal binding information of the virtual digital human model, the model mesh can be divided into several basic sub-regions according to the influence range of the skeletal nodes; based on the semantic functions of each sub-region in human-computer interaction (such as facial expression communication, gesture interaction, body posture, etc.), adjacent sub-regions with similar semantic functions are merged into functional regions. For example, sub-regions closely related to facial expression communication, such as eyes, lips, and eyebrows, are merged into facial expression regions; sub-regions related to gesture interaction, such as palms and fingers, are merged into hand regions; and sub-regions responsible for displaying body posture, such as the torso and limbs, are respectively classified into torso regions and limb regions. Through the above two-level division method based on skeletal topology and semantic functions, the virtual digital human model can be divided into multiple regions.

[0022] In some embodiments of this disclosure, initial weights (i.e., visual weights) for each region of the virtual digital human model can be established through statistical analysis based on eye-tracking data collected from actual scenarios. Specifically, eye-tracking data of multiple test users interacting with the virtual digital human can be collected in a preset user test scenario, recording the gaze duration and number of gazes on each region of the virtual digital human model. For each region, the proportion of the average gaze duration to the total gaze duration P_t and the proportion of the average number of gazes to the total number of gazes P_f are calculated. The two are weighted and summed to obtain the visual attention score of the region S=α×P_t+β×P_f, where α and β are preset weight coefficients and α+β=1 (e.g., α=0.6, β=0.4). The visual attention scores of all regions are normalized and mapped to the [0,1] interval to obtain the initial weight W=S / S_max for each region, where S_max is the maximum visual attention score in all regions. As an example, the virtual digital human model can be divided into a high-priority region (weight 0.7-1.0), a medium-priority region (weight 0.4-0.7), and a low-priority region (weight 0.1-0.4).

[0023] The initial weights for high-priority areas may include: eyes - weight 1.0, lips - weight 0.95, eyebrows - weight 0.9, and hands - weight 0.85.

[0024] Medium priority areas may include: cheeks - weight 0.68, nose - weight 0.65, forehead - weight 0.6, neck - weight 0.58, bangs - weight 0.55, shoulders - weight 0.5, and hairline - weight 0.45.

[0025] Low-priority areas may include: abdomen - weight 0.35, back - weight 0.3, forearm - weight 0.25, calf - weight 0.2, back of head - weight 0.18, hair ends - weight 0.15, and non-critical clothing parts - weight 0.1.

[0026] Step 102: Update the initial weights of each region according to the interaction parameters to obtain the first weight of each region.

[0027] Optionally, the interaction parameters include at least one of the following: interaction state, distance between the user and the interaction, and display area ratio of each region.

[0028] The interaction state can include a speaking state and a gesture state. Let the preset adjustment range for the interaction state be Δ_state, where the adjustment range for the speaking state is Δ_speak = 0.2, and the adjustment range for the gesture state is Δ_gesture = 0.3. When the virtual digital human model is in a speaking state, the weight of the mouth region in the virtual digital human model is increased based on the initial weight W_mouth, i.e., the updated weight = W_mouth + Δ_speak. For example, if the initial weight of the lips is 0.95, the updated weight in the speaking state is 0.95 + 0.2 = 1.0 (with a truncated upper limit of 1.0). When the virtual digital human model is in a gesture state, the weight of the hand region in the virtual digital human model is increased based on the initial weight W_hand, i.e., the updated weight = W_hand + Δ_gesture. For example, if the initial weight of the hand is 0.85, the updated weight in the gesture state is 0.85 + 0.3 = 1.0 (with a truncated upper limit of 1.0). The aforementioned adjustment ranges Δ_speak and Δ_gesture can be preset based on the degree of influence of each interaction state on visual attention in the actual application scenario. The updated weight values ​​are restricted to the [0,1] interval by a truncation function.

[0029] Updating the initial weights of each region based on the interaction distance can include: Let the interaction distance be d, the maximum distance threshold be d_max (e.g., 2 meters), the minimum distance threshold be d_min (e.g., 0.5 meters), and the preset maximum adjustment range be Δ_max (e.g., 0.3). When the interaction distance d is greater than or equal to the maximum distance threshold d_max, the reduction range Δ_down = Δ_max × min((d - d_max) / (d_max), 1) is calculated. The facial region weights are reduced from the initial weights in the virtual digital human model, i.e., updated weight = initial weight - Δ_down. For example, when d = 2 meters, Δ_down = 0.3 × min((2-2) / 2, 1) = 0; when d = 3 meters, Δ_down = 0.3 × min((3-2) / 2, 1) = 0.15; when d ≥ 4 meters, Δ_down = 0.3. When the interaction distance d is less than or equal to the minimum distance threshold d_min, the increase Δ_up = Δ_max × min((d_min - d) / d_min, 1) is calculated. This increases the facial region weight based on the initial weight in the virtual digital human model; that is, the updated weight = initial weight + Δ_up. For example, when d = 0.5 meters, Δ_up = 0.3 × min((0.5 - 0.5) / 0.5, 1) = 0; when d = 0.3 meters, Δ_up = 0.3 × min((0.5 - 0.3) / 0.5, 1) = 0.12. The updated weight value is restricted to the interval [0, 1] by a truncation function.

[0030] The initial weights of each region are updated based on its display area percentage. This can be achieved by: Let R be the display area percentage of a region, and R_min be the minimum percentage threshold (e.g., 5%). Within the virtual digital human model, a first target region is identified where the display area percentage R is less than or equal to the minimum percentage threshold R_min. A decay coefficient k = R / R_min is calculated, and the weight of this first target region is reduced from its initial weight. That is, the updated weight = initial weight × k. For example, when the display area percentage R = 5%, k = 5% / 5% = 1, and the weight remains unchanged; when R = 2.5%, k = 2.5% / 5% = 0.5, and the updated weight = initial weight × 0.5. The updated weight value is then restricted to the [0,1] interval using a truncation function.

[0031] The initial weights of each region are updated based on their location. This can include dividing the screen into a central region and an edge region. Let the normalized distance from the center point of a region to the center of the screen be D_c (D_c∈[0,1], where 0 represents the center of the screen and 1 represents the edge), and let the preset position adjustment increment be Δ_pos (e.g., 0.2). From the multiple regions of the virtual digital human model, a third target region is determined that displays in the central region of the screen (i.e., D_c is less than or equal to the central region threshold D_th, e.g., D_th=0.3). The position gain factor G_pos=(D_th-D_c) / D_th is calculated, and the weight of the third target region is increased based on its initial weight. That is, the updated weight = initial weight + Δ_pos×G_pos. For example, when a region is located in the center of the screen (D_c=0), G_pos=1, and the updated weight = initial weight + 0.2; when D_c=0.15, G_pos=0.5, and the updated weight = initial weight + 0.1. The updated weight values ​​are restricted to the [0,1] interval by a truncation function.

[0032] Step 103: Based on the first weight and the preset mapping rules, select the initial LOD model corresponding to each region from the LOD models of multiple accuracy levels.

[0033] It should be noted that the weights of each region are positively correlated with the accuracy level of the LOD model; in other words, the higher the region weight, the higher the accuracy level, and consequently, the higher the accuracy. Taking three accuracy levels of LOD models as an example: LOD0 is a high-accuracy model: 8000 polygons + 2048 textures + SSS material + SSAO; LOD1 is a medium-accuracy model: 3000 polygons + 1024 textures + standard PBR; and LOD2 is a low-accuracy model: 1000 polygons + 512 textures + simplified lighting. Preset mapping rules may include: If the region weight is ≥0.8, select the high-precision level model LOD0; If 0.8 > region weight ≥ 0.4, select the medium precision level model LOD1; If the region weight is less than 0.4, select the low-precision level model LOD2.

[0034] In some embodiments of this disclosure, the interaction parameters may further include the user's eye-tracking data. Updating the initial weights of each region based on the interaction parameters may further include: acquiring the user's eye-tracking data to determine the user's gaze point on the virtual digital human model; adjusting the first weight of each region a second time based on the region where the gaze point is located to obtain a second weight; subsequently, based on the second weight and a preset mapping rule, reselecting the initial LOD model corresponding to each region, or adjusting the threshold for determining the accuracy level of subsequent LOD models based on the second weight.

[0035] As an example, real-time images of the user's eyes can be captured via a camera. The direction of the gaze can be analyzed using an eye model and corneal reflection technology at the center of the pupil. Combined with the three-dimensional spatial mapping between the screen coordinate system and the virtual digital human model, the precise landing point area of ​​the user's gaze on the surface of the virtual digital human model can be calculated. Let the current weight of the region where the gaze point is located be W_cur, the preset base value of the gaze gain be Δ_gaze (e.g., 0.15), the gaze duration be T_gaze, and the preset time threshold be T_th (e.g., 0.5 seconds). If the gaze point falls on a certain region, the gaze gain Δ = Δ_gaze × min(T_gaze / T_th, 2) is calculated. The weight is then increased based on the existing first weight for that region, i.e., the updated weight = W_cur + Δ. For example, if the fixation point falls on the mouth area and the fixation duration is 0.5 seconds, then Δ = 0.15 × min(0.5 / 0.5, 2) = 0.15, and the weight of the mouth area increases by 0.15; if the fixation duration is 1 second, then Δ = 0.15 × min(1 / 0.5, 2) = 0.3. Meanwhile, for areas that have not been fixated on for a long time (i.e., the unfixed time exceeds the preset forgetting threshold T_forget, for example, 2 seconds), their weight can be appropriately reduced. The decay coefficient λ = max(1 - 0.1 × (T_ungazed - T_forget) / T_forget, 0.5) is calculated, and the updated weight = W_cur × λ. For example, when a certain area has not been fixated on for 4 seconds, λ = max(1 - 0.1 × (4 - 2) / 2, 0.5) = 0.9, and the weight decays to 0.9 times the original value. The updated weight value is restricted to the interval [0,1] by a truncation function.

[0036] By introducing eye-tracking data, this disclosure can capture the dynamic distribution of user attention in real time, enabling the allocation of rendering resources to not only depend on macro-interaction scenarios (such as speech and gestures) and physical parameters (distance and display area), but also to more accurately match the user's immediate visual focus. This achieves a more granular real-time perception of user attention, allowing the adjustment of rendering weights to go from the scene level to the foveation level. Thus, with limited hardware resources, the highest rendering precision is given to the local area that the user is currently most concerned with, further enhancing the visual realism and user experience of the virtual digital human, while avoiding resource misallocation caused by fixed rules.

[0037] Step 104: Determine the CPU load and memory usage of the virtual digital human model based on the initial LOD model corresponding to each region.

[0038] Step 105: Adjust the LOD model accuracy level for each region based on CPU load and / or memory usage, and determine the LOD model for each region after adjustment.

[0039] When the CPU load is greater than or equal to the maximum load threshold (e.g., 80%), the accuracy level of the LOD model for each region can be reduced based on the initial LOD model for each region, thereby ensuring the frame rate. When the CPU load is less than or equal to the minimum load threshold (e.g., 50%), the accuracy level of the LOD model for each region can be improved based on the initial LOD model for each region.

[0040] When memory usage is greater than or equal to the highest memory threshold (e.g., 450MB), a second target region with a first weight less than the first weight threshold (e.g., 0.3) is identified among multiple regions. Based on the initial LOD model corresponding to the second target region, the accuracy level of the LOD model corresponding to the second target region is reduced.

[0041] Step 106: Render the corresponding regions of the virtual digital human model based on the adjusted LOD models for each region.

[0042] In some embodiments of this disclosure, when LOD model switching is required in each region, a 0.3-second fade-in / fade-out transition can be used, that is, the new model fades in while the old model fades out.

[0043] This embodiment introduces interactive parameters to dynamically update the rendering weights of each region and initially selects a Level of Detail (LOD) model based on these weights, achieving an initial match between rendering accuracy and visual attention. Based on the estimated CPU and memory load of the initially selected model, the LOD accuracy is calibrated a second time, thus incorporating real-time system performance constraints into the decision-making loop. Under conditions of limited overall resources, there is no need to sacrifice model rendering quality. Instead, a dual dynamic adjustment is made based on the current interactive scenario and the actual hardware load, automatically and rationally allocating resources to prioritize the rendering quality of high-attention areas while suppressing resource consumption in non-critical areas. This effectively improves the overall visual expressiveness of the virtual digital human while maintaining stable and smooth system operation.

[0044] During rendering, loading all high-resolution textures at once would exceed memory limits. Therefore, in some embodiments of this disclosure, an on-demand loading strategy is proposed, prioritizing the loading of regions with higher weight. As an example, the texture loading strategy may include: 1. Initial loading phase: All textures use 128x128 thumbnails (for quick display); Total size < 10MB, loading time < 1 second; The user immediately sees the virtual human (low-quality version); 2. Progressive loading phase: Sort the weights from high to low and load high-resolution textures sequentially according to the priority queue; After each image is loaded, immediately replace the texture of the corresponding part; Use a fade-in effect (0.3 seconds) for a smooth transition.

[0045] To better understand the embodiments of this disclosure, the following provides a rendering process example: Scenario: A user has a video conversation with a virtual digital human (camera distance 1 meter, frontal view). [Initialization Phase - Seconds 0-1]: 1. Load 128×128 thumbnail textures of all parts of the virtual digital human model (total <10MB); 2. Display a low-poly version of the virtual character (total 3000 faces, LOD2); 3. Start the background texture streaming loading queue; 4. What the user sees: a blurry but complete virtual human (loading time < 1 second); [Progressive loading phase - seconds 1-3]: 5. Load the 2048×2048 ASTC compressed texture map of the face (2.7MB) according to priority; 6. Switch the face to LOD0 high-poly model (8000 polygons); 7. Enable facial SSS material and SSAO; 8. The user sees: the face is clear, but the body is still blurry; 9. Load a 1024×1024 texture of the hand in the background (0.7MB); 10. Switch the hand to LOD1 medium model (3000 faces / hand); 11. Users see: clear facial and hand details; 12. Load 1024×1024 textures for the torso and hair in the background (0.7MB each); 13. Switch the torso and hair to LOD1 mid-model (3000 and 2000 polygons respectively); 14. The user sees: the upper body is clear, but the legs are still blurry; 15. Load 512×512 textures of limbs in the background (0.3MB); 16. The limbs are kept in LOD2 low-poly form (1000 faces). 17. Users see: a complete, high-quality virtual human; [Stable operation phase - after 3 seconds]: 18. Performance monitoring: FPS=58, GPU=65%, Memory=280MB → Normal; 19. Face BlendShape updates at 60fps; 20. The torso skeleton updates at 30fps; 21. Limb bones are updated at 15fps (60fps animation generated via interpolation). [Interaction state change - 10 seconds]: 22. When a virtual human makes a waving gesture, the hand priority weight increases by 0.3. 23. Switch the hand from LOD1 to LOD0 high-poly model (1500 polygons / hand); 24. Enable hand shadow projection onto face (256×256 shadow map); 25. The hand skeleton update frequency remains at 60fps; 26. What the user sees: a natural and fluid waving motion, with the shadow of the hand projected onto the face; [Camera zooms out - 20 seconds]: 27. Camera distance changed to 3 meters → Face priority weight -0.2; 28. Reduce facial textures to 1024×1024 (to free up memory); 29. Disable SSAO, only retain pre-calculated AO; 30. The face model was switched from LOD0 to LOD1 (3000 polygons + 1024 textures + standard PBR). 31. Performance monitoring: FPS=60, GPU=50%, Memory=220MB → Excellent; [Performance fluctuation handling - 30 seconds]: 32. Suppose the user opens other tabs, and GPU resources are being used; 33. FPS dropped to 45 → triggering downgrade strategy; 34. The limbs are switched to LOD2 low-poly model (already LOD2, no change); 35. Switch the torso to LOD2 low-poly model (3000 polygons → 1000 polygons); 36. Shadow map resolution reduced to 256×256; 37. Disable real-time lighting in non-facial areas; 38. Maintain LOD0 and 2048 textures for the face (priority required); 39. FPS restored to 55 → stable operation; 40. Users see: facial quality remains unchanged, body is slightly blurry, but still smooth; [Camera close-up - 40 seconds]: 41. Camera distance changes to 0.3 meters (close-up of face) → Facial weight +0.3; 42. Facial textures have been increased to 2048×2048 (if previously downgraded); 43. Enable high-precision normal mapping; 44. Enable SSS subsurface scattering effect; 45. Face BlendShape updates at 60fps; 46. ​​Other parts off-screen will automatically be downgraded to LOD2; 47. Users will see: extremely delicate facial details and realistic skin texture.

[0046] Figure 2 This is a schematic diagram of a virtual digital human adaptive rendering device provided in an embodiment of this disclosure. Figure 2 As shown, the virtual digital human adaptive rendering device may include: an acquisition module 201, a weight update module 202, a first determination module 203, a second determination module 204, an adjustment module 205, and a rendering module 206.

[0047] The acquisition module 201 is used to acquire the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model.

[0048] The weight update module 202 is used to update the initial weights of each region according to the interaction parameters to obtain the first weight of each region.

[0049] The first determining module 203 is used to select the initial LOD model corresponding to each region from multiple LOD models of precision levels according to the first weight and the preset mapping rules.

[0050] The second determining module 204 is used to determine the CPU load and memory usage of the virtual digital human model based on the initial LOD model corresponding to each region.

[0051] The adjustment module 205 is used to adjust the accuracy level of the LOD model corresponding to each region based on CPU load and / or memory usage, and to determine the LOD model corresponding to each region after adjustment.

[0052] The rendering module 206 is used to render the corresponding areas of the virtual digital human model based on the adjusted LOD models of each area.

[0053] In some embodiments of this disclosure, the interaction parameters include at least one of the following: interaction state, distance between the user and the user, and display area ratio of each region.

[0054] In some embodiments of this disclosure, the interaction state includes a speaking state and a gesture state; the weight update module 202 is specifically used to: increase the weight of the mouth region based on the initial weight of the mouth region in the virtual digital human model in response to the virtual digital human model being in a speaking state; and increase the weight of the hand region based on the initial weight of the hand region in the virtual digital human model in response to the virtual digital human model being in a gesture state.

[0055] In some embodiments of this disclosure, the weight update module 202 is specifically configured to: reduce the weight of the facial region in the virtual digital human model based on the initial weight of the facial region in response to the distance between the user and the interactive user being greater than or equal to the maximum distance threshold; and increase the weight of the facial region in the virtual digital human model based on the initial weight of the facial region in response to the distance between the user and the interactive user being less than or equal to the minimum distance threshold.

[0056] In some embodiments of this disclosure, the weight update module 202 is specifically used to: determine a first target region whose display area ratio is less than or equal to a minimum ratio threshold among multiple regions; and reduce the weight of the first target region based on the initial weight of the first target region.

[0057] In some embodiments of this disclosure, the accuracy level and accuracy are positively correlated. The adjustment module 205 is specifically used to: reduce the accuracy level of the LOD model corresponding to each region based on the initial LOD model corresponding to each region in response to the CPU load being greater than or equal to the highest load threshold; and increase the accuracy level of the LOD model corresponding to each region based on the initial LOD model corresponding to each region in response to the CPU load being less than or equal to the lowest load threshold.

[0058] In some embodiments of this disclosure, the accuracy level and accuracy are positively correlated. The adjustment module 205 is specifically used to: in response to memory usage being greater than or equal to the highest memory threshold, determine a second target region among multiple regions whose first weight is less than the first weight threshold; and reduce the accuracy level of the LOD model corresponding to the second target region based on the initial LOD model corresponding to the second target region.

[0059] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0060] To implement the above embodiments, this disclosure also proposes an electronic device, including: a processor and a memory communicatively connected to the processor; the memory stores computer execution instructions; the processor executes the computer execution instructions stored in the memory to implement the method provided in the foregoing embodiments.

[0061] To implement the above embodiments, this disclosure also proposes a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the methods provided in the foregoing embodiments.

[0062] To implement the above embodiments, this disclosure also proposes a computer program product, including a computer program that, when executed by a processor, implements the methods provided in the foregoing embodiments.

[0063] In the foregoing descriptions of the embodiments, the terms "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this disclosure. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0064] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this disclosure, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0065] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing custom logic functions or processes, and the scope of preferred embodiments of this disclosure includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of this disclosure pertain.

[0066] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-included system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of computer-readable media include: an electrical connection having one or more wires (electronic device), a portable computer disk drive (magnetic device), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Alternatively, the computer-readable medium may be paper or other suitable media on which the program can be printed, since the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in a computer memory.

[0067] It should be understood that various parts of this disclosure can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0068] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0069] Furthermore, the functional units in the various embodiments of this disclosure can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium.

[0070] The storage medium mentioned above can be a read-only memory, a disk, or an optical disk, etc. Although embodiments of the present disclosure have been shown and described above, it is to be understood that the above embodiments are exemplary and should not be construed as limiting the present disclosure. Those skilled in the art can make changes, modifications, substitutions, and variations to the above embodiments within the scope of the present disclosure.

Claims

1. A virtual digital human adaptive rendering method, characterized in that, Includes the following steps: Obtain the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model; The initial weights of each region are updated according to the interaction parameters to obtain the first weight of each region; Based on the first weight and the preset mapping rules, the initial LOD model corresponding to each region is selected from multiple LOD models of precision levels; The CPU load and memory usage of the virtual digital human model are determined based on the initial LOD model corresponding to each region. Based on the CPU load and / or memory usage, adjust the LOD model accuracy level corresponding to each region, and determine the LOD model corresponding to each region after adjustment; Based on the adjusted LOD models corresponding to each region, the corresponding regions of the virtual digital human model are rendered.

2. The method according to claim 1, characterized in that, The interaction parameters include at least one of the following: interaction status, distance between the user and the user, and display area ratio of each region.

3. The method according to claim 2, characterized in that, The interaction states include speaking state and gesture state; the initial weights of each region are updated according to the interaction states, including: In response to the virtual digital human model being in a speaking state, the weight of the mouth region in the virtual digital human model is increased based on the initial weight of the mouth region; In response to the virtual digital human model being in a gesture state, the weight of the hand region in the virtual digital human model is increased based on the initial weight of the hand region.

4. The method according to claim 2, characterized in that, The initial weights of each region are updated based on the distance to the interacting user, including: In response to the distance between the user and the interacting user being greater than or equal to the maximum distance threshold, the facial region weight in the virtual digital human model is reduced based on the initial weight of the facial region. In response to the distance between the user and the interacting user being less than or equal to a minimum distance threshold, the facial region weights in the virtual digital human model are increased based on the initial weights of the facial region.

5. The method according to claim 2, characterized in that, The initial weights of each region are updated based on the display area ratio of each region, including: A first target region is identified among the plurality of regions whose display area percentage is less than or equal to a minimum percentage threshold. The weight of the first target region is reduced based on its initial weight.

6. The method according to claim 1, characterized in that, The accuracy level and accuracy are positively correlated. Adjusting the LOD model accuracy level for each region based on the CPU load includes: In response to the CPU load being greater than or equal to the maximum load threshold, the accuracy level of the LOD model corresponding to each region is reduced based on the initial LOD model corresponding to each region. In response to the CPU load being less than or equal to the minimum load threshold, the accuracy level of the LOD model corresponding to each region is increased based on the initial LOD model corresponding to each region.

7. The method according to claim 2, characterized in that, The accuracy level and accuracy are positively correlated. Adjusting the LOD model accuracy level for each region based on the memory usage includes: In response to the memory usage being greater than or equal to the highest memory threshold, a second target region in which the first weight is less than the first weight threshold is determined from the plurality of regions; Based on the initial LOD model corresponding to the second target region, the accuracy level of the LOD model corresponding to the second target region is reduced.

8. A virtual digital human adaptive rendering device, characterized in that, include: The acquisition module is used to acquire the interaction parameters of the virtual digital human model to be rendered, as well as the initial weights of each region in the multiple regions divided by the virtual digital human model. The weight update module is used to update the initial weight of each region according to the interaction parameters to obtain the first weight of each region; The first determining module is used to select the initial LOD model corresponding to each region from multiple LOD models of precision levels according to the first weight and the preset mapping rules. The second determining module is used to determine the CPU load and memory usage of the virtual digital human model based on the initial LOD model corresponding to each region; The adjustment module is used to adjust the accuracy level of the LOD model corresponding to each region based on the CPU load and / or the memory usage, and to determine the LOD model corresponding to each region after adjustment; The rendering module is used to render the corresponding areas of the virtual digital human model based on the adjusted LOD models corresponding to each area.

9. An electronic device, characterized in that, include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the method as described in any one of claims 1-7.

Images