Neural Video Compression Using Key Frames and Keypoints

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Streaming video content places high demands on bandwidth and often compromises video quality due to inconsistent connection quality, with traditional compression solutions providing marginal savings.

Innovation Solution

Utilizing neural networks to compress and decompress video data by transmitting key frames and keypoints, allowing for efficient reconstruction of video frames with improved quality and reduced bandwidth requirements.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If traditional video compression solutions are employed, then bandwidth requirements are reduced, but video quality deteriorates

Engineering Contradiction:

Improvebandwidth requirementsVSAvoidvideo quality

Core Design Contradiction:

Quantity of substanceVSManufacturing precision

Solution Approach 1:

The patent segments video frames into key frames and key points, transmitting only the essential components rather than complete frames. This segmentation allows for significant bandwidth reduction while maintaining video quality, as the receiver can reconstruct full frames from the transmitted key points using neural networks.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The patent introduces neural networks as intermediaries in the compression and decompression process. The sender neural network extracts key points from frames, and the receiver neural network reconstructs frames from these key points. This intermediary processing enables high-quality video transmission with reduced bandwidth requirements.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of energy

If traditional video compression solutions are employed, then some bandwidth savings are achieved, but connection quality remains compromised due to inconsistent network infrastructure

Engineering Contradiction:

Improvebandwidth savingsVSAvoidconnection quality

Core Design Contradiction:

Loss of energyVSReliability

Solution Approach 1:

The patent changes the fundamental parameters of video compression by transitioning from traditional block-based compression to neural network-based key point extraction. This parameter change enables the system to adapt to varying network conditions while maintaining connection quality, as the human visual system is more sensitive to key structural points than to pixel-level details.

Inventive Principle:

Principle #35Parameter changes

3Manufacturing precision

If more video data is transmitted to maintain quality, then video quality is preserved, but bandwidth consumption increases

Engineering Contradiction:

Improvevideo qualityVSAvoidbandwidth consumption

Core Design Contradiction:

Manufacturing precisionVSQuantity of substance

Solution Approach 1:

The patent extracts only the most essential visual information (key points and key frames) from video frames for transmission. By taking out only the critical structural elements that define video content rather than transmitting all pixel data, the system maintains video quality while dramatically reducing bandwidth consumption.

Inventive Principle:

Principle #2Taking out (Extraction)

Data Source

PatentUS12506886B2Video compression using neural networks

Publication Date: 2025.12.23 NVIDIA CORP

AI summary

Apparatuses, systems, and techniques to perform compression of video data using neural networks to facilitate video streaming, such as video conferencing. In at least one embodiment, a sender transmits to a receiver a key frame from video data and one or more keypoints identified by a neural network from said video data, and a receiver reconstructs video data using said key frame and one or more received keypoints.