Paper2Fig100k

该数据集由来自研究论文中的图-文对组成，用于训练和验证文本到图形生成模型。该数据集对于训练潜在扩散模型至关重要，这种模型能够根据文本描述生成科学图表。数据集的规模包括81,194个训练样本和21,259个验证样本，其任务专注于文本到图形的生成。

This dataset consists of graph-text pairs sourced from research papers, and is utilized for training and validating text-to-graph generation models. It is critical for training latent diffusion models, which are capable of generating scientific diagrams based on textual descriptions. The dataset includes 81,194 training samples and 21,259 validation samples, with its task focusing on text-to-graph generation.

搜集汇总

背景与挑战

背景概述

OCR-VQGAN是一个专注于生成包含清晰文本的图像编码器，其核心创新是引入了OCR感知损失函数。配套的Paper2Fig100k数据集包含10万+科研论文图表，特别适合文本-图像生成任务的研究。