Intelligent ai generated video guided by hardware encoder
The system addresses high computational costs and frame inconsistencies in video generation by using AI to generate key frames and hardware encoding for remaining frames, achieving efficient and coherent video output with reduced latency and improved visual quality.
Patent Information
- Application Number
- PCT/CN2024/074500
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-01-29
- Publication Date
- 2025-08-07
AI Technical Summary
Existing video generation technologies face challenges such as high computational costs, incoherent frames, artifacts, and flickering due to the application of image-to-image generation techniques to each frame of a video, leading to restricted output resolution, duration, and increased latency.
A system that uses an AI model to generate modified key frames based on selected key frames and a hardware encoder to encode remaining frames using motion information, bypassing the AI model for the remaining frames, combined with lightweight post-processing to refine interpolated frames.
This approach reduces computational costs, enhances video resolution and duration, and improves coherence and stability by eliminating artifacts and flickering, resulting in high-quality, efficient video generation.
Smart Images

Figure CN2024074500_07082025_PF_FP_ABST
Abstract
Description
INTELLIGENT AI GENERATED VIDEO GUIDED BY HARDWARE ENCODERBACKGROUND
[0001] In general, the creation of artificial intelligence generated content (AIGC) may involve the use of artificial intelligence (AI) to assist or replace manual content generation by generating content based on user-provided keywords or requirements (e.g., contextual information) . For example, image-to-image generation might use a pre-trained generative model (e.g., stable diffusion, DALL-E 2) to transform an input still image into a stylized or enhanced output image based on contextual information provided by the user. Attempts to apply image-to-image generation techniques to video, however, face several challenges. For example, the computational costs of applying image-to-image generation to each decoded frame of an input video are substantial (e.g., resulting in restricted output resolution and duration, as well as greater latency) . Additionally, the generated video sequence typically contains incoherent, jarring frames and artifacts, unpredictable content, a loss of continuity, and / or inconsistency and flickering across frames.BRIEF DESCRIPTION OF THE DRAWINGS
[0002] The various advantages of the embodiments will become apparent to one skilled in the art by reading the following specification and appended claims, and by referencing the following drawings, in which:
[0003] FIG. 1 is a block diagram of an example of an enhanced video-to-video artificial intelligence generated content (AIGC) system according to an embodiment;
[0004] FIG. 2 is an illustration of an example of a video-to-video AIGC process flow according to an embodiment;
[0005] FIG. 3 is an illustration of an example of a transformation from frames in an input video to frames in an output video according to an embodiment;
[0006] FIG. 4 is a flowchart of an example of a method of operating a performance-enhanced computing system according to an embodiment;
[0007] FIG. 5 is a block diagram of an example of a performance-enhanced computing system according to an embodiment; and
[0008] FIG. 6 is an illustration of an example of a semiconductor package apparatus according to an embodiment.DETAILED DESCRIPTION
[0009] Turning now to FIG. 1, an enhanced video-to-video artificial intelligence generated content (AIGC) system 10 is shown in which an artificial intelligence (AI) model 12 (e.g., generative adversarial network variant) generates modified key frames 14 based on selected key frames in an input video 16. For example, the modified key frames 14 can be stylized or enhanced versions of corresponding key frames in the input video 16, wherein the key frames are selected in response to scene change conditions and / or interval conditions detected in the input video 16. The AIGC system 10 also includes a hardware encoder 18 that encodes remaining frames 20 in the input video 16 based on motion information (e.g., encoded bitstream motion vectors that facilitate interpolation) . In the illustrated example, a modified output video 22 includes the modified key frames 14 and the remaining frames 20.
[0010] As will be discussed in greater detail, the remaining frames 20 bypass the AI model 12, which substantially reduces computational costs. Indeed, the reduced computational costs enables the resolution and duration of the modified output video 22 to be greater (e.g., longer than a short video clip) , as well as the latency associated with generating the modified output video 22 to be reduced. Additionally, the speed and accuracy of the hardware encoder 18 eliminates incoherent, jarring frames and artifacts, unpredictable content, continuity losses, and inconsistency and flickering across frames from the modified output video 22.
[0011] FIG. 2 shows a process flow 30 (30a-30i) in which a first operation 30a decodes the input video 16 into a sequence of frames 32. The first operation 30a may involve extracting individual RGB (red, green, blue) frames from a compressed video file. A second operation 30b intelligently selects sparse key frames 34 (34a, 34i) based on scene cuts, user-defined intervals, and so forth. More details on the second operation 30b are articulated below. A third operation 30c conducts image-to-image AIGC generation for the sparse key frames 34 to obtain modified key frames 14 (14a, 14i, e.g., painting style rather than photograph) . In an embodiment, the third operation 30c uses the AI model 12, which can be a pre-trained generative model that includes an adapter 36 and stable diffusion 38 (e.g., seed value generator) . Accordingly, the AI model 12 can draw inferences based on semantic / contextual information 40 and content 42 extracted from the sparse key frames 34.
[0012] A fourth operation 30d encodes all AI-generated key frames 14 as I-frames 44 (44a, 44i, e.g., intra prediction only) . A fifth operation 30e re-encodes the original sequence of frames 32, designating the selected sparse key frames 34 as I-frames and remaining frames 20 as P-frames (e.g., with inter-prediction mode enabled) . With continuing reference to FIGs. 2 and 3, a sixth operation 30f substitutes all I-frames in the re-encoded bitstream with the I-frames 44 (e.g., encoded AI generated key frames) from the fourth operation 30d. The illustrated sixth operation 30f also retains the P-frame content to reuse motion information.
[0013] A seventh operation 30g decodes the manipulated bitstream, wherein an eighth operation 30h refines decoded P-frames via tailored lightweight post-processing to reduce artifacts. Additionally, a ninth operation 30i re-encodes the refined sequences to generate the modified output video 22 (e.g., final output) . Thus, selective AI synthesis combined with motion information reuse, substitution, and lightweight post-processing enables substantially (e.g., 10x) faster high-quality AIGC video-to-video generation versus existing approaches.
[0014] Key Frame Selection
[0015] The second operation 30b (e.g., key frame selection stage) identifies sparse but useful frames for AI generation. Given input video and interval thresholds, the second operation 30b selects sparse key frames 34 using one or more of scene cut detection (e.g., via histogram comparison between frames, with large differences indicating cuts, pixel difference thresholds also detecting potential cuts and / or frames meeting either threshold) or a user-defined maximum interval between key frames 34. The user-defined maximum interval between key frames 34 triggers periodic selection. If the interval between previously selected key frames 34 exceeds this threshold, a new key frame 34 is selected. Accordingly, key frames 34 are intelligently selected in response to abrupt scene changes detected by histogram and pixel analysis, or when intervals between frames exceed pre-set limits. Such an approach minimizes required AI synthesis while maximizing content reuse.
[0016] Motion Information Reuse from Encoding Engine
[0017] The fifth and sixth operations 30e, 30f leverage a highly optimized encoding engine through rapid and precise motion estimation. In video-to-video generation, the motion of the generated video closely aligns with the original. The technology described herein capitalizes on this alignment by utilizing the encoding hardware to efficiently extract motion information between frames.
[0018] Specifically, the encoded bitstream motion vectors from the optimized encoding engine are reused to perform zero-cost, seamless interpolation between the sparsely generated AI generated key frames 14. This motion-based interpolation eliminates the need for costly intermediate AI synthesis.
[0019] Post-processing
[0020] Additionally, lightweight tailored post-processing removes potential artifacts in the interpolated frames that can arise from the sparse key frame 14 encoding. This post-processing includes addressing inconsistencies from the limited AI-generated anchor / key frames 14.
[0021] Thus, by tapping into the speed and accuracy of a hardware-accelerated encoder for motion data, the technology described herein achieves high-quality interpolation without expensive frame-by-frame AI generation. The encoding engine performs the heavy lifting while the lightweight post-processing “polishes” the interpolated frames. This optimization strategy maximizes efficiency and quality.
[0022] More particularly, the eighth operation 30h (e.g., post-processing stage) refines decoded P-frames between the AI-generated key frames 14 in the manipulated bitstream. Since P-frames are reconstructed from embedded motion information, the quality of the P-frames significantly impacts the final output along with the AI-generated key frames 14. Post-processing addresses artifacts caused by the limited intra blocks in P-frames lacking motion vectors, as well as unanticipated deformations from the AI model 12. By enhancing the reconstructed P-frames between key frames 14, post-processing optimizes overall quality after AI generation and motion-based interpolation.
[0023] Meanwhile, the relatively expensive deep learning in this flanking stage is bypassed to conserve more system resources for the critical and heavy AIGC key frame generation tasks. In one example, three approaches are introduced in post-processing to enhance the visual quality of the P-frame:
[0024] - “Blockiness” reduction: For each P-frame, a blockiness metric is computed by analyzing discrete cosine transform (DCT) coefficient distributions to quantify blocking artifact severity. Iterative deblocking can be applied with an adaptive filtering strength based on the blockiness level to reduce blocking while preserving edges.
[0025] - Flicker reduction: For each P-frame, a flicker metric is computed by analyzing histogram differences between the current frame and temporally adjacent frames. A Gaussian filter may then be applied along motion trajectories when flicker is detected.
[0026] - Shot Change adjustment: For each P-frame, shot / scene change is estimated by comparing color histograms with key frame using, for example, a chi-square metric (e.g., measure of the difference between the observed and expected frequencies of the outcomes of a set of events or variables) . For gradual shot changes, histogram normalization may be applied for color coherency. For relatively fast transitions, histogram matching can be applied to key frames 14 for consistency. Additionally, contrast may be locally adjusted based on gradient histogram analysis.
[0027] Accordingly, tailored analysis identifies defects in P-frames. Targeted filtering and adjustments repair identified issues between AI key frames 14, enhancing overall quality.
[0028] Employing sparse AI key frames 14 and motion information reuse on top of highly optimized encoding engine provides significant improvements in terms of efficiency and visual quality compared to prior end-to-end learned models. The combination of selective AI generation, motion information reuse, substitution, and lightweight post-processing enables faster high-quality AIGC video-to-video generation versus conventional approaches. Additionally, visual quality improves substantially. For example, interpolation and compositing guided by semantic masks heighten continuity and stability. Foreground-background separation with asymmetric rates lessens flickering.
[0029] FIG. 4 shows a method 50 of operating a performance-enhanced computing system. The method 50 may generally be implemented in a video-to-video AIGC system such as, for example, the enhanced video-to-video AIGC system 10 (FIG. 1) , already discussed. More particularly, the method 50 may be implemented in one or more modules as a set of logic instructions stored in a machine-or computer-readable storage medium such as random access memory (RAM) , read only memory (ROM) , programmable ROM (PROM) , firmware, flash memory, etc., in hardware, or any combination thereof. For example, hardware implementations may include configurable logic, fixed-functionality logic, or any combination thereof. Examples of configurable logic (e.g., configurable hardware) include suitably configured programmable logic arrays (PLAs) , field programmable gate arrays (FPGAs) , complex programmable logic devices (CPLDs) , and general purpose microprocessors. Examples of fixed-functionality logic (e.g., fixed-functionality hardware) include suitably configured application specific integrated circuits (ASICs) , combinational logic circuits, and sequential logic circuits. The configurable or fixed- functionality logic can be implemented with complementary metal oxide semiconductor (CMOS) logic circuits, transistor-transistor logic (TTL) logic circuits, or other circuits.
[0030] For example, computer program code to carry out operations shown in the method 50 can be written in any combination of one or more programming languages, including an object oriented programming language such as JAVA, SMALLTALK, C++ or the like and conventional procedural programming languages, such as the “C” programming language or similar programming languages. Additionally, logic instructions might include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, state-setting data, configuration data for integrated circuitry, state information that personalizes electronic circuitry and / or other structural components that are native to hardware (e.g., host processor, central processing unit / CPU, microcontroller, etc. ) .
[0031] Illustrated processing block 52 provides for generating, via an AI model (e.g., pre-trained generative model) , modified key frames based on selected key frames in an input video. In one example, block 52 identifies one or more of the selected key frames in response to a scene change condition. In such a case, the scene change condition may be detected based on histogram data and / or pixel difference threshold data. Block 52 may also identify one or more of the selected key frames in response to an interval (e.g., maximum interval between key frames) condition.
[0032] Block 54 encodes, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames bypass the AI model. In an embodiment, block 54 encodes the selected key frames as I-frames, wherein the remaining frames are encoded as P-frames. Block 56 may conduct a post-processing adjustment on the remaining frames, wherein the post-processing adjustment includes one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.
[0033] The method 50 therefore enhances performance at least to the extent that bypassing the remaining frames around the AI model substantially reduces computational costs. Indeed, the reduced computational costs enables the resolution and duration of the modified output video to be greater, as well as the latency associated with generating the modified output video to be reduced. Additionally, the speed and accuracy of the hardware encoder eliminates incoherent, jarring frames and artifacts, unpredictable content, continuity losses, and inconsistency and flickering across frames from the modified output video.
[0034] Turning now to FIG. 5, a performance-enhanced computing system 280 is shown. The system 280 may generally be part of an electronic device / platform having computing functionality (e.g., personal digital assistant / PDA, notebook computer, tablet computer, convertible tablet, server) , communications functionality (e.g., smart phone) , imaging functionality (e.g., camera, camcorder) , media playing functionality (e.g., smart television / TV) , wearable functionality (e.g., watch, eyewear, headwear, footwear, jewelry) , vehicular functionality (e.g., car, truck, motorcycle) , robotic functionality (e.g., autonomous robot) , Internet of Things (IoT) functionality, drone functionality, etc., or any combination thereof.
[0035] In the illustrated example, the system 280 includes a host processor 282 (e.g., CPU) having an integrated memory controller (IMC) 284 that is coupled to a system memory 286 (e.g., dual inline memory module / DIMM) . In an embodiment, an IO (input / output) module 288 is coupled to the host processor 282. The illustrated IO module 288 communicates with, for example, a display 290 (e.g., touch screen, liquid crystal display / LCD, light emitting diode / LED display) , and a network controller 292 (e.g., wired and / or wireless) . The host processor 282 may be combined with the IO module 288, a graphics processor 294, and an AI accelerator 296 into a system on chip (SoC) 298.
[0036] In an embodiment, the graphics processor 294 includes logic 300 (e.g., including executable program instructions, configurable hardware, fixed-functionality hardware, etc., or any combination thereof) to perform one or more aspects of the method 50 (FIG. 4) and / or the process flow 30 (FIG. 2) , already discussed. Thus, the logic 300 can generate, via an AI model, modified key frames based on selected key frames in an input video. The logic 300 can also encode, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames are to bypass the AI model.
[0037] The computing system 280 is therefore considered performance-enhanced at least to the extent that bypassing the remaining frames around the AI model substantially reduces computational costs. Indeed, the reduced computational costs enables the resolution and duration of the modified output video to be greater, as well as the latency associated with generating the modified output video to be reduced. Additionally, the speed and accuracy of the hardware encoder eliminates incoherent, jarring frames and artifacts, unpredictable content, continuity losses, and inconsistency and flickering across frames from the modified output video.
[0038] FIG. 6 shows a semiconductor apparatus 350 (e.g., chip, die, package) . The illustrated apparatus 350 includes one or more substrates 352 (e.g., silicon, sapphire, gallium arsenide) and logic 354 (e.g., transistor array and other integrated circuit / IC components) coupled to the substrate (s) 352. In an embodiment, the logic 354 includes an AI model 356 and a hardware encoder 358, and implements one or more aspects of method 50 (FIG. 4) and / or the process flow 30 (FIG. 2) , already discussed. The logic 354 may also be substituted for the logic 300 (FIG. 5) , already discussed.
[0039] The logic 354 can be implemented at least partly in configurable or fixed-functionality hardware. In one example, the logic 354 includes transistor channel regions that are positioned (e.g., embedded) within the substrate (s) 352. Thus, the interface between the logic 354 and the substrate (s) 352 may not be an abrupt junction. The logic 354 may also be considered to include an epitaxial layer that is grown on an initial wafer of the substrate (s) 352.
[0040] Additional Notes and Examples:
[0041] Example 1 includes a performance-enhanced computing system comprising a network controller and a processor coupled to the network controller, wherein the processor includes logic coupled to one or more substrates, the logic to generate, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video and encode, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames are to bypass the AI model.
[0042] Example 2 includes the computing system of Example 1, wherein the logic is further to conduct a post-processing adjustment on the remaining frames, and wherein the post-processing adjustment is to include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.
[0043] Example 3 includes the computing system of Example 1, wherein the logic is further to identify one or more of the selected key frames in response to a scene change condition.
[0044] Example 4 includes the computing system of Example 3, wherein the scene change condition is detected based on histogram data.
[0045] Example 5 includes the computing system of Example 3, wherein the scene change condition is detected based on pixel difference threshold data.
[0046] Example 6 includes the computing system of Example 1, wherein the logic is further to identify one or more of the selected key frames in response to an interval condition.
[0047] Example 7 includes the computing system of any one of Examples 1 to 6, wherein the logic is further to encode, via the hardware encoder, the selected key frames as I-frames, and wherein the remaining frames are encoded as P-frames.
[0048] Example 8 includes a semiconductor apparatus comprising one or more substrates, and logic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable or fixed-functionality hardware, the logic to generate, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video, and encode, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames are to bypass the AI model.
[0049] Example 9 includes the semiconductor apparatus of Example 8, wherein the logic is further to conduct a post-processing adjustment on the remaining frames, and wherein the post-processing adjustment is to include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.
[0050] Example 10 includes the semiconductor apparatus of Example 8, wherein the logic is further to identify one or more of the selected key frames in response to a scene change condition.
[0051] Example 11 includes the semiconductor apparatus of Example 10, wherein the scene change condition is detected based on histogram data.
[0052] Example 12 includes the semiconductor apparatus of Example 10, wherein the scene change condition is detected based on pixel difference threshold data.
[0053] Example 13 includes the semiconductor apparatus of Example 8, wherein the logic is further to identify one or more of the selected key frames in response to an interval condition.
[0054] Example 14 includes the semiconductor apparatus of any one of Examples 8 to 13, wherein the logic is further to encode, via the hardware encoder, the selected key frames as I-frames, and wherein the remaining frames are encoded as P-frames.
[0055] Example 15 includes the semiconductor apparatus of any one of Examples 8 to 13, wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.
[0056] Example 16 includes a method of operating a performance-enhanced computing system, the method comprising generating, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video, and encoding, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames bypass the AI model.
[0057] Example 17 includes the method of Example 16, further including conducting a post-processing adjustment on the remaining frames, wherein the post-processing adjustment include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.
[0058] Example 18 includes the method of Example 16, further including identifying one or more of the selected key frames in response to a scene change condition, wherein the scene change condition is detected based on one or more of histogram data or pixel difference threshold data.
[0059] Example 19 includes the method of Example 16, further including identifying one or more of the selected key frames in response to an interval condition.
[0060] Example 20 includes the method of any one of Examples 16 to 19, further including encoding, via the hardware encoder, the selected key frames as I-frames, wherein the remaining frames are encoded as P-frames.
[0061] Example 21 includes an apparatus comprising means for performing the method of any one of Examples 16 to 20.
[0062] In general, the video generation technology described herein video employs AI-generated key frames, foreground-background decomposition, and intelligent frame interpolation to efficiently synthesize high-quality, realistic videos with increased generation speed boost and enhanced coherence and stability compared to end-to-end learned models. More particularly, the technology described herein generates coherent, realistic videos that maintain sequential logic, spatial relations, and global consistency across hundreds or even thousands of frames. The technology involves sparse key frame generation integrated with light-weight motion-based interpolation. Specifically, embodiments generate sparse key frames with targeted content using an AI model, reducing synthesis requirements. Embodiments also re-encode the original video sequence and substitute sparse key frames with encoded AI-generated frames while retaining non-key frames to reuse the embedded motion information in the bitstream. In addition, embodiments applying post-processing to refine interpolation quality.
[0063] The technology described herein enables efficient, scalable synthesis of customized high-fidelity video content, unlocking new applications in areas such as media production, computer generated animation, AI assistants, and so forth. The combination of sparse AI key frames and motion information reuse with a highly optimized encoding engine leads to considerable increases in generation efficiency over existing end-to-end models. Moreover, the faster video generation is not accompanied by the cost of degraded visuals. In fact, human assessments and automated measures both show noticeable visual enhancements with the technology described herein compared to other end-to-end models. The videos contain substantially fewer flickering artifacts and greater continuity between frames, yielding higher overall realism and coherence. Objective metrics quantifying inter-frame consistency (e.g., inter-frame variation metrics) are improved significantly.
[0064] Embodiments are applicable for use with all types of semiconductor integrated circuit ( “IC” ) chips. Examples of these IC chips include but are not limited to processors, controllers, chipset components, programmable logic arrays (PLAs) , memory chips, network chips, systems on chip (SoCs) , SSD / NAND controller ASICs, and the like. In addition, in some of the drawings, signal conductor lines are represented with lines. Some may be different, to indicate more constituent signal paths, have a number label, to indicate a number of constituent signal paths, and / or have arrows at one or more ends, to indicate primary information flow direction. This, however, should not be construed in a limiting manner. Rather, such added detail may be used in connection with one or more exemplary embodiments to facilitate easier understanding of a circuit. Any represented signal lines, whether or not having additional information, may actually comprise one or more signals that may travel in multiple directions and may be implemented with any suitable type of signal scheme, e.g., digital or analog lines implemented with differential pairs, optical fiber lines, and / or single-ended lines.
[0065] Example sizes / models / values / ranges may have been given, although embodiments are not limited to the same. As manufacturing techniques (e.g., photolithography) mature over time, it is expected that devices of smaller size could be manufactured. In addition, well known power / ground connections to IC chips and other components may or may not be shown within the figures, for simplicity of illustration and discussion, and so as not to obscure certain aspects of the embodiments. Further, arrangements may be shown in block diagram form in order to avoid obscuring embodiments, and also in view of the fact that specifics with respect to implementation of such block diagram arrangements are highly dependent upon the computing system within which the embodiment is to be implemented, i.e., such specifics should be well within purview of one skilled in the art. Where specific details (e.g., circuits) are set forth in order to describe example embodiments, it should be apparent to one skilled in the art that embodiments can be practiced without, or with variation of, these specific details. The description is thus to be regarded as illustrative instead of limiting.
[0066] The term “coupled” may be used herein to refer to any type of relationship, direct or indirect, between the components in question, and may apply to electrical, mechanical, fluid, optical, electromagnetic, electromechanical or other connections. In addition, the terms “first” , “second” , etc. may be used herein only to facilitate discussion, and carry no particular temporal or chronological significance unless otherwise indicated.
[0067] As used in this application and in the claims, a list of items joined by the term “one or more of” may mean any combination of the listed terms. For example, the phrases “one or more of A, B or C” may mean A; B; C; A and B; A and C; B and C; or A, B and C.
[0068] Those skilled in the art will appreciate from the foregoing description that the broad techniques of the embodiments can be implemented in a variety of forms. Therefore, while the embodiments have been described in connection with particular examples thereof, the true scope of the embodiments should not be so limited since other modifications will become apparent to the skilled practitioner upon a study of the drawings, specification, and following claims.
Claims
1.A performance-enhanced computing system comprising:a network controller; anda processor coupled to the network controller, wherein the processor includes logic coupled to one or more substrates, the logic to:generate, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video, andencode, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames are to bypass the AI model.2.The computing system of claim 1, wherein the logic is further to conduct a post-processing adjustment on the remaining frames, and wherein the post-processing adjustment is to include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.3.The computing system of claim 1, wherein the logic is further to identify one or more of the selected key frames in response to a scene change condition.4.The computing system of claim 3, wherein the scene change condition is detected based on histogram data.5.The computing system of claim 3, wherein the scene change condition is detected based on pixel difference threshold data.6.The computing system of claim 1, wherein the logic is further to identify one or more of the selected key frames in response to an interval condition.7.The computing system of any one of claims 1 to 6, wherein the logic is further to encode, via the hardware encoder, the selected key frames as I-frames, and wherein the remaining frames are encoded as P-frames.8.A semiconductor apparatus comprising:one or more substrates; andlogic coupled to the one or more substrates, wherein the logic is implemented at least partly in one or more of configurable or fixed-functionality hardware, the logic to:generate, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video; andencode, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames are to bypass the AI model.9.The semiconductor apparatus of claim 8, wherein the logic is further to conduct a post-processing adjustment on the remaining frames, and wherein the post-processing adjustment is to include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.10.The semiconductor apparatus of claim 8, wherein the logic is further to identify one or more of the selected key frames in response to a scene change condition.11.The semiconductor apparatus of claim 10, wherein the scene change condition is detected based on histogram data.12.The semiconductor apparatus of claim 10, wherein the scene change condition is detected based on pixel difference threshold data.13.The semiconductor apparatus of claim 8, wherein the logic is further to identify one or more of the selected key frames in response to an interval condition.14.The semiconductor apparatus of any one of claims 8 to 13, wherein the logic is further to encode, via the hardware encoder, the selected key frames as I-frames, and wherein the remaining frames are encoded as P-frames.15.The semiconductor apparatus of any one of claims 8 to 13, wherein the logic coupled to the one or more substrates includes transistor channel regions that are positioned within the one or more substrates.16.A method of operating a performance-enhanced computing system, the method comprising:generating, via an artificial intelligence (AI) model, modified key frames based on selected key frames in an input video; andencoding, via a hardware encoder, remaining frames in the input video based on motion information, wherein the remaining frames bypass the AI model.17.The method of claim 16, further including conducting a post-processing adjustment on the remaining frames, wherein the post-processing adjustment include one or more of a blockiness reduction, a flicker reduction or a shot change adjustment.18.The method of claim 16, further including identifying one or more of the selected key frames in response to a scene change condition, wherein the scene change condition is detected based on one or more of histogram data or pixel difference threshold data.19.The method of claim 16, further including identifying one or more of the selected key frames in response to an interval condition.20.The method of any one of claims 16 to 19, further including encoding, via the hardware encoder, the selected key frames as I-frames, wherein the remaining frames are encoded as P-frames.
Citation Information
Patent Citations
Ai-assisted programmable hardware video codec
CN113874916A
Video coding and decoding processing method and device, computer equipment and storage medium
CN116233445A
Online training-based encoder tuning with multi model selection in neural image compression
WO2023192096A1