Temporal-Spatial Discriminator GANs for Efficient Video Generation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing generative adversarial networks face challenges in efficiently generating high-resolution and long-duration videos due to the heavy computational requirements and complexity of processing large datasets, particularly in tasks requiring accurate temporal and spatial coherence.

Innovation Solution

The implementation of a Dual Video Discriminator GAN (DVD-GAN) with a spatial discriminator and a temporal discriminator, applying separate downsampling to reduce the resolution of input images, combined with a factorized self-attention mechanism to reduce memory footprint and enhance processing efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a single discriminator network processes full-resolution video sequences, then temporal and spatial coherence can be evaluated, but computational complexity and processing time increase significantly

Engineering Contradiction:
Improvetemporal and spatial coherence evaluationVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The discriminator network is segmented into two specialized components: a spatial discriminator that evaluates spatial coherence at reduced temporal resolution, and a temporal discriminator that evaluates temporal coherence at reduced spatial resolution. This segmentation allows each discriminator to focus on specific aspects of video quality without processing the full computational burden of high-resolution complete video sequences, thereby maintaining evaluation precision while reducing overall computational complexity.

Inventive Principle:
Principle #1Segmentation

2Manufacturing precision

If high-resolution video sequences are processed in full detail, then generation fidelity improves, but processing speed and training efficiency decrease

Engineering Contradiction:
Improvevideo generation fidelityVSAvoidprocessing speed
Core Design Contradiction:
Manufacturing precisionVSProductivity

Solution Approach 1:

Different resolution levels are applied to different discriminators based on their specific evaluation needs. The spatial discriminator receives video sequences with reduced temporal resolution but maintains spatial detail for accurate spatial feature evaluation. The temporal discriminator receives video sequences with reduced spatial resolution but maintains temporal sequence integrity for accurate temporal feature evaluation. This local quality approach ensures that each discriminator operates at the optimal resolution for its specific function, maintaining overall generation fidelity while improving processing speed.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3973459B1Generative adversarial networks with temporal and spatial discriminators for efficient video generation
Publication Date: 2025.07.02 GDM HOLDING LLC
  • EP3973459B1 patent drawingFigure 1
  • EP3973459B1 patent drawingFigure 2
  • EP3973459B1 patent drawingFigure 3

AI summary

The present disclosure proposes the use of a dual discriminator network that comprises a temporal discriminator network for discriminating based on temporal features of a series of images and a spatial discriminator network for discriminating based on spatial features of individual images. The training methods described herein provide improvements in computational efficiency. This is achieved by applying the spatial discriminator network to a set of one or more images that have reduced temporal resolution and applying the temporal discriminator network to a set of images that have reduced spatial resolution. This allows each of the discriminator networks to be applied more efficiently in order to produce a discriminator score for use in training the generator, whilst maintaining accuracy of the discriminator network. In addition, this allows a generator network to be trained to more accurately generate sequences of images, through the use of the improved discriminator.