Deep Learning-based Video Super Resolution Accelerator IP
Streaming video accounts for more than 80% of internet traffic, and over 90% of content is streamed in FHD or lower resolutions f…
Overview
Streaming video accounts for more than 80% of internet traffic, and over 90% of content is streamed in FHD or lower resolutions from a variety of services such as OTT and cloud gaming. On the other hand, the market share of displays that support 4K or higher exceeds 50%. This is much higher than the number of high-resolution content being delivered. It's only natural that service users want higher quality video for a better viewing experience on their devices.
The media industry archives a lot of content that was created before high-resolution displays were available on the market. For example, media & entertainment companies own legacy media content that was produced in lower resolutions, and cloud gaming services have classic (retro) games.
Converting outdated or unserved contents to modern display standards can turn it into a new revenue opportunity.
DeepField-SR is a fixed functional hardware accelerator IP for FPGA and ASIC, offering the highest computational efficiency for Video Super Resolution. Based on proprietary AI models trained with real world video dataset and fusing spatio-temporal information in multiple frames, DeepField-SR produces superior high resolution video quality and upscales to fit lager displays.
Bluedot Hardware IP Solves Trade-off
Achieving both high-quality upgrading and real-time processing with Software AI model accelerated by GPU is highly challenging. Typically, one must to choose between the two. However, BLUEDOT overcomes this trade-off by accelerating AI models with dedicated hardware IP.
Upscaling video for transcoding
By leveraging an encoding-aware proprietary upscaling technique at a lower bitrate, the solution reduces bit usage compared to the conventional upscaling methods while preserving high visual quality
Key features
In-house AI models
- Denoising (e.g. compression artifact)
- Upscaling with generating visual details
- Utilizing spatial-temporal features from consecutive frames for natural and seamless video enhancement
- Weights for Networks are updatable
Fully Hardware IP
- Upscaling FHD to 4K at 120fps at 1.0GHz in real-time
- Utilize built-in resizer with AI Super Resolution for fractional scaling ratio (e.g. 2.5x)
- 8-/10-bit depth input video supported
- AMBA AXI4 compatible
Target devices
- ASIC
Block Diagram
Files
Note: some files may require an NDA depending on provider policy.
Specifications
Identity
Provider
Learn more about Video Processing IP core
Configurable Processors for Video Processing SOCs
FPGA-based video surveillance comes of age
Picking the right MPSoC-based video architecture: Part 1
Video encoding with low-cost FPGAs for multi-channel H.264 surveillance
Analysis: ARC's Configurable Video Subsystems
Frequently asked questions about Video Processing IP
What is Deep Learning-based Video Super Resolution Accelerator IP?
Deep Learning-based Video Super Resolution Accelerator IP is a Video Processing IP core from BLUEDOT listed on Semi IP Hub.
How should engineers evaluate this Video Processing?
Engineers should review the overview, key features, supported foundries and nodes, maturity, deliverables, and provider information before shortlisting this Video Processing IP.
Can this semiconductor IP be compared with similar products?
Yes. Buyers can compare this product with similar semiconductor IP cores or IP families based on category, provider, process options, and structured technical specifications.