Differentiable Mesh Tiling for Seamless Non-Square Image Generation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional tiling systems lack flexibility, accuracy, and computational efficiency when creating tile-able graphical objects, often resulting in images with gaps or background regions and requiring computationally slow optimization techniques.

Innovation Solution

A differentiable tiling system that utilizes dynamic edge weights and a differentiable renderer to generate a textured 2D mesh that satisfies overlap and tile-able boundary conditions, optimizing both geometry and texture to produce a plausible, tile-able image from a textual input, using gradient-based optimization and a trained image diffusion model.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If conventional image manipulation methods are used to create tile-able images, then the process is simple, but the flexibility and accuracy in manipulating non-square object geometry and appearance is poor

Engineering Contradiction:

Improveflexibility in manipulating non-square object geometryVSAvoidcomplexity of image manipulation process

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent replaces traditional mechanical image manipulation methods with a neural network-based system. The neural network automatically learns and applies geometric transformations and texture mappings to generate tile-able images from non-square source images, eliminating the need for complex manual manipulation while achieving superior flexibility and accuracy in geometry and appearance control.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system changes key parameters including the aspect ratio of source images (allowing non-square inputs), geometric transformation parameters (scaling, rotating, skewing), and texture mapping parameters. By optimizing these parameters through the neural network, the system achieves flexible manipulation of non-square objects while maintaining tile-ability constraints.

Inventive Principle:

Principle #35Parameter changes

2Manufacturing precision

If conventional methods are used to ensure visual consistency in tile-able images, then computational resources are saved, but the accuracy and visual quality of the tile-able output is poor

Engineering Contradiction:

Improveaccuracy of tile-able image generationVSAvoidcomputational efficiency

Core Design Contradiction:

Manufacturing precisionVSUse of energy by stationary object

Solution Approach 1:

The system performs preliminary geometric transformations and texture mappings on the source image before generating the final tile-able output. By pre-processing the image with appropriate geometric adjustments and texture assignments, the neural network can focus computational resources on optimizing the tile-ability constraints and visual consistency, achieving high accuracy without excessive computational overhead.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The system generates multiple candidate tile-able images by copying and transforming the source image in different ways (different geometric transformations, texture mappings). The neural network then selects and refines the best candidate that satisfies tile-ability constraints and visual consistency requirements, improving accuracy through comparative evaluation rather than single-pass generation.

Inventive Principle:

Principle #26Copying

3Reliability

If complex neural network models are used to optimize geometry and texture, then the quality of tile-able images improves, but the training time and computational cost increase

Engineering Contradiction:

Improvequality and plausibility of tile-able designVSAvoidtraining time of neural network

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The neural network is divided into separate modules: a geometric transformation module that handles spatial transformations, and a texture mapping module that handles appearance and color adjustments. This segmentation allows each module to be trained independently on specific tasks, reducing overall training time while maintaining high quality output through specialized optimization in each module.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS12633006B2Generating tile-able images utilizing a differentiable mesh generation and rendering pipeline

Publication Date: 2026.05.19 ADOBE INC

AI summary

The present disclosure is directed toward systems, methods, and non-transitory computer readable media that provide a differentiable tiling system that generates aesthetically plausible, periodic, and tile-able non-square imagery using machine learning and a text-guided, fully automatic generative approach. Namely, given a textual description of the object and a symmetry pattern of the 2D plane, the system produces a textured 2D mesh which visually resembles the textual description, adheres to the geometric rules which ensure it can be used to tile the plane, and contains only the foreground object. Indeed, the disclosed systems generate a plausible textured 2D triangular mesh that visually matches the textual input and optimizes both the texture and the shape of the mesh and satisfy an overlap condition and a tile-able condition. Using the described methods, the differentiable tiling system generates the mesh such that the edges and the vertices align between repeatable instances of the mesh.