Machine Learning Texture Compression for GPU Compatibility

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Current texture compression methods in video games result in large storage requirements and slow loading times due to fixed compression ratios, with high-compression formats being incompatible with graphics processing units (GPUs).

Innovation Solution

A method using trained machine learning models, such as generative adversarial networks (GANs), to convert graphics hardware-incompatible compressed textures into hardware-compatible formats directly usable by the GPU, allowing for higher compression ratios without compromising image quality.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If high-compression texture formats are used, then storage space is reduced, but compatibility with GPU is lost

Engineering Contradiction:

Improvestorage spaceVSAvoidGPU compatibility

Core Design Contradiction:

Quantity of substanceVSAdaptability or versatility

Solution Approach 1:

The patent introduces a machine learning model as an intermediary converter that translates between compression formats. The model takes GPU-incompatible high-compression textures as input and outputs GPU-compatible textures in standard formats (BC7, BC6, BC5, BC4, or DXT), enabling the system to use high-compression storage while maintaining GPU compatibility through automated format conversion.

Inventive Principle:

Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent performs preliminary conversion of texture formats before they reach the GPU. By pre-processing textures through the machine learning model to convert them into GPU-compatible formats, the system eliminates the need for runtime format conversion, ensuring both storage efficiency and hardware compatibility are achieved in advance.

Inventive Principle:

Principle #10Preliminary action

2Loss of energy

If block compression is used at runtime, then memory and bandwidth are saved, but compression ratio is limited

Engineering Contradiction:

Improvememory and bandwidthVSAvoidcompression ratio

Core Design Contradiction:

Loss of energyVSProductivity

Solution Approach 1:

The patent changes the compression parameters by applying machine learning-based compression algorithms that achieve superior compression ratios compared to traditional block compression. The system uses trained models to optimize compression settings, delivering textures at higher compression ratios while maintaining quality and GPU compatibility, thus reducing memory and bandwidth usage more effectively.

Inventive Principle:

Principle #35Parameter changes

3Adaptability or versatility

If CPU decompression and re-compression is performed, then GPU compatibility is achieved, but processing time increases

Engineering Contradiction:

ImproveGPU compatibilityVSAvoidprocessing time

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The patent replaces the traditional CPU-based decompression and re-compression mechanical process with a machine learning model that performs direct format conversion. This substitution eliminates the two-step CPU process (decompress to RGB(A) then re-compress to BC/DXT), reducing processing time while achieving the same GPU compatibility goal through a more efficient single-step conversion.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3776218B1Machine learning applied to textures compression or upscaling

Publication Date: 2023.03.29 MICROSOFT TECHNOLOGY LICENSING LLC

Figure 1
Figure 2A
Figure 2B~2C

AI summary

Methods and devices for generating hardware compatible compressed textures may include accessing, at runtime of an application program, graphics hardware incompatible compressed textures in a format incompatible with a graphics processing unit (GPU). The methods and devices may include converting the graphics hardware incompatible compressed textures directly into hardware compatible compressed textures usable by the GPU using a trained machine learning model.