Generative Adversarial Network for Document Layout Generation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing technologies lack effective methods for learning and generating layouts in documents, with insufficient discussion on the learning process and room for improvement.
Innovation Solution
A learning system utilizing Generative Adversarial Networks that extracts constraints from inputted layouts to generate constrained training data, learns to generate layouts using random numbers, and discriminates between generated and constrained training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional layout generation methods are used, then layout generation is possible, but learning effectiveness is insufficient
Solution Approach 1:
The patent applies preliminary action by pre-extracting constraints from layout data and pre-generating constrained training data before the actual layout generation task. The training data is prepared in advance with extracted constraints (such as column count, row count, element positions) so that when layout generation is needed, the system can directly use this pre-processed data for efficient learning and generation, resolving the contradiction between learning effectiveness and generation efficiency.
Solution Approach 2:
The patent segments the layout generation process into distinct functional components: constraint extraction module, training data generation module, and layout generation module. Each module handles a specific aspect of the task independently, allowing the constraint extraction to be performed separately on historical data, then the generated constraints are used during the actual generation process. This segmentation enables both thorough learning from historical data and efficient real-time generation.
2Measurement precision
If constraints are extracted from layouts, then learning precision is improved, but processing complexity increases
Solution Approach 1:
The patent applies the extraction principle by isolating the constraint information from the full layout data. Instead of processing the entire complex layout structure, the system extracts only the essential constraints (such as grid dimensions, element positions, and layout patterns) and uses these simplified constraint representations for training. This extraction reduces the complexity of subsequent processing while maintaining high precision in the learned patterns.
Solution Approach 2:
The patent uses copying by creating a simplified representation of the layout constraints as training data. The original complex layout data is copied and transformed into a constrained training format that contains only the essential constraint information needed for learning. This copied constraint data can be processed more efficiently while preserving the essential patterns, reducing processing complexity without sacrificing learning precision.
Data Source
AI summary
A learning system includes: a training data generation unit that extracts a constraint of an inputted layout and that generates constrained training data; and a learning unit that learns, by using Generative Adversarial Networks, a learning unit that generates a generated layout by using a random number and a layout discrimination unit that discriminates the generated layout and the constrained training data. According to such a learning system, it is possible to properly learn the layout in a document or the like.