How Video Analytic Helps to Power Broadcasting Business u Jin Huang u CTO u Arcvideo Inc.
2 Agenda • Arcvideo introduction • How GPU been used in media processing pipeline • How video analytic helps for broadcasting business • Summary
3 What is Arcvideo • Arcvideo is our software defined Video Solution, including Broadcasting Level Codec engines, Intelligent Video Analytic engines, End Device Player and Cloud Video Services • Core products are video solution cover ingesting, production, distribution and playback stage, handling high quality/performance Video Transcoding, Video Processing, Intelligent video analyzing, Video Streaming and Cloud Services • Arcvideo Solutions are used to many areas, like TV Operators, Broadcasters, Content Providers, Telecom, Enterprise, Education and others.
4 TV Station Telecom Cable TV Phone Operator Operator Media Companies Mobile Cloud Pad PC Operator Operator UGC STB Enterprise Multi-Screen Monitoring Transcoding Game Smart User Device APP Video Interaction & H5 CMS Analytic Security Multi-CDN Content User Data DRM Education & QoS Operation Mining Codec Video Analytic Device APP Cloud
5 GPU Advantages for Video Industry -Multiple resources to accelerate video pipeline • Hardware accelerated decoding and encoding • Less Servers and Space, save up to 80% • Fast delivery time, up to 85% faster – Fast Decoding capability with good error resilience • Less Power consumption, save up to 60% – NVENC for multiple sessions of encoding, with various quality level and latency mode choices • CUDA accelerated video post processing – Adaptive Deinterlacing/Frame Rate Up-conversion – Various video enhancement algorithms • Deep Learning acceleration for intelligent = video analytic One ArcVideo – Face Recognition GPU Server – Object Recognition
6 Arcvideo Customized parts: • • CUDA accelerated video Codec Transcoding pipeline optimization – Seamless buffer sharing between HW Decoding, Video – MPEG2 422/444, 10bit, HDR Processing/Analyzing and HW Encoding – Apple ProRes ／ SMPTE VC3 – Handling various streaming content dynamic change – Perceptual Based Coding – Reduce unnecessary overhead moving uncompressed • Deep Learning acceleration buffer – Real time video quality enhancement • Modified NVENC with customization of QP – Face Recognition – Better rate control over NVENC – Object Recognition, like Car and Cloth • CUDA accelerated video processing and video • User interaction for live event broadcasting analytic – Real time VR stitching and AR rendering – Scaling/Video composition/CC/Subtitle
7 What Arcvideo could benefit • Impressive video performance comparing to traditional CPU or HW multimedia solutions: – Good Hardware accelerated Decoding/Encoding performance, and tons of CUDA cores • Highly customizable via CUDA programming – Easy to customize CUDA accelerated video post processing and video analytic algorithms – Flexible CUDA programming to easily fit customer request in very short time • Balanced GPU and CUDA core configuration – Both Tesla and GRID provide various combination of GPU and CUDA core to fit different user scenarios – Mature server vendors ecosystem to find reliable GPU servers, depends on task burden, pick multiple GPU board and achieve highest density
How Video Analytic helps-Video Encoding -Perceptual Video Coding • Improve encoding video quality with same bitrate • Video scene learning and detection to adjust encoding parameters adaptively • Working with traditional video encoding scheme, perceptualvideo coding can enhance compression efficiency • HVS (human visual system) characteristicshelp to exploit perceptual redundancy and improve video compression efficiency, but involves more computing – Luminance, Contrast sensitivity, fovea, etc – SSIM/M-SSIM/CW-SSIM/VIF/VQM – Perceptual Noise from Spatial and Temporal Perspectives – Region of Interest
How Video Analytic helps-Video Quality Enhance -Deep Learning for Video Quality Improvement • Improve video quality for OTT limited bitrate scenarios • CNN training – Extract highly detailed, rich color and texture blocks from high quality video frames automatically as candidate training set – Get a low resolution version as input, and original version as output – Using lots of GPU acceleration to make it real time
How Video Analytic helps-HDR • EOTF/OETF, Color Space conversion between SDR and HDR or different HDR schemes – HLG, PQ, HDR10, S-Gamut, Philips/Technicolor HDR – S-Log3, BT.709, BT.2020 – No standard way to do the conversion, proprietary tone expansion different in quality – Better Tone Mapping and Inverse Tone Mapping visual effect • HDR characteristics help to improve encoding quality – Color fidelity, Adaptive Gamma Curve
How Video Analytic helps-Workflow Automation -Improve the efficiency of workflow • Improve efficiency of Content Production and Management by reducing manual workloads • Media Management – Smart metadata extraction based on Face recognition, Object detection and recognition, Scene detection • Content Production – Auto channel recognition for better UX – Intelligent scene detection and video segmenting for quick editing – Fast video segment detection, including AD, Opening, Ending using video/audio fingerprint for AD Insertion and replacing – Speech to Text for auto-subtitle/cc
12 Summary: Why use GPU and How -To solve computing and understanding problem
13 Global Leader in Multimedia Solution
Feedback • Email: jhuang@arcvideo.com • LinkedIn: huangjin.hz@gmail.com • Please complete the Presenter Evaluation sent to you by email or through the GTC Mobile App. Your feedback is important!
Recommend
More recommend
