Video Resolution Upscaling With Flow-Guided Temporal Consistency

Solution Overview

Problem

Conventional video resolution techniques for upsampling digital videos suffer from visual artifacts such as blurriness and temporal flickering, failing to achieve visually pleasing high-resolution videos.

Innovation Solution

Employing a video resolution system that utilizes a flow guided feature propagation module, anti-aliasing blocks, and a high-frequency shuttle within a generative adversarial network (GAN) to enhance temporal consistency and add fine-grained details to video frames.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Manufacturing precision

If conventional upsampling techniques are used to increase video resolution, then the resolution of video frames is improved, but visual artifacts such as blurriness and temporal flickering occur

Engineering Contradiction:

Improvevideo resolutionVSAvoidvisual artifacts

Core Design Contradiction:

Manufacturing precisionVSObject-affected harmful factors

Solution Approach 1:

The patent introduces flow-guided feature propagation as an intermediary mechanism between the input low-resolution video and the output high-resolution video. Optical flow fields are computed to guide the propagation of features across frames, enabling the system to transfer high-frequency details while maintaining temporal consistency. This intermediary guidance mechanism prevents the direct upsampling artifacts that cause blurriness and flickering.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent performs preliminary computation of optical flow fields and feature extractions before the actual upsampling process. By pre-computing the flow-guided feature propagation maps from multiple frames, the system prepares the high-frequency detail information in advance, which is then applied during upsampling to avoid temporal inconsistencies and flickering artifacts in the final output.

Inventive Principle:

Principle #10Preliminary action

2Manufacturing precision

If conventional upsampling techniques are used to increase video resolution, then the resolution of video frames is improved, but temporal consistency between frames deteriorates

Engineering Contradiction:

Improvevideo resolutionVSAvoidtemporal consistency

Core Design Contradiction:

Manufacturing precisionVSStability of the object's composition

Solution Approach 1:

The patent introduces flow-guided feature propagation as an intermediary mechanism between the input low-resolution video and the output high-resolution video. Optical flow fields are computed to guide the propagation of features across frames, enabling the system to transfer high-frequency details while maintaining temporal consistency. This intermediary guidance mechanism prevents the direct upsampling artifacts that cause blurriness and flickering.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The system uses feedback from multiple frames through optical flow computation to guide the upsampling process. By continuously referencing the motion fields and feature correspondences across frames, the system adjusts the feature propagation to maintain temporal consistency, ensuring that high-frequency details are consistently reconstructed across all frames without flickering.

Inventive Principle:

Principle #23Feedback

3Stability of the object's composition

If machine learning models with flow guided feature propagation and anti-aliasing blocks are used, then temporal consistency and high-frequency details are improved, but device complexity increases

Engineering Contradiction:

Improvetemporal consistencyVSAvoidprocessing system complexity

Core Design Contradiction:

Stability of the object's compositionVSDevice complexity

Solution Approach 1:

The patent segments the video processing task into distinct functional modules: optical flow computation module, feature extraction module, flow-guided feature propagation module, anti-aliasing block, and upsampling module. Each module performs a specific function, making the overall complex system more manageable and allowing for optimized implementation of each component independently while maintaining temporal consistency.

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS20260057476A1Machine learning video resolution adjustment

Publication Date: 2026.02.26 ADOBE INC

AI summary

Machine learning video resolution adjustment techniques are described. An input digital video is received having a plurality of frames and processed by one or more machine-learning models using a processing device. The processing is performed such that the input digital video having frames in a first resolution is adjusted into an output digital video having the frames in a second resolution. Examples of processing include use of a flow guided feature propagation module, anti-aliasing blocks, and/or a high-frequency shuttle.