synthetic-seismic-vlm
收藏资源简介:
Synthetic Seismic VLM是一个合成地震多模态问答数据集,专为视觉问答、图像分割和图像到文本等任务设计。数据集包含1行数据,每行数据由多个字段构成,具体包括:原始地震图像序列(images)、分割掩码图像序列(masks)、任务指令(instruction)、问题文本(question)、可选的推理或描述文本(reason)、答案文本(answer)、支持性文本证据的JSON字符串(evidence)以及包含边界框、类别、颜色等区域元数据的JSON字符串(regions)。该数据集旨在提供证据基础的多模态问答样本,适用于地震图像分析、视觉语言模型训练等相关研究与应用场景。数据集语言为英语,规模属于小样本类别(n<1K),采用MIT许可证。
Synthetic Seismic VLM is a synthetic seismic multimodal question-answering dataset specifically designed for tasks including visual question answering, image segmentation, and image-to-text. The dataset contains 1 row of data, with each row consisting of multiple fields: original seismic image sequences (images), segmented mask image sequences (masks), task instruction (instruction), question text (question), optional reasoning or descriptive text (reason), answer text (answer), JSON string of supporting textual evidence (evidence), and JSON string containing regional metadata such as bounding boxes, categories, and colors (regions). This dataset aims to provide evidence-based multimodal question-answering samples, applicable to research and application scenarios such as seismic image analysis, visual language model training and other related fields. The dataset uses English, has a small sample size (n < 1000), and is released under the MIT License.
数据集概述:Synthetic Seismic VLM
- 许可证:MIT
- 任务类别:视觉问答、图像分割、图像到文本
- 语言:英语
- 数据规模:少于1,000条(共527条)
数据内容
该数据集包含合成地震多模态问答样本,具体包括:
- 原始地震图像 (
images) - 分割掩码图像 (
masks) - 任务指令 (
instruction) - 问题文本 (
question) - 可选推理/描述文本 (
reason) - 答案文本 (
answer) - 支持文本证据 (
evidence):JSON字符串格式 - 区域元数据 (
regions):JSON字符串格式,包含边界框、类别、颜色信息
仓库地址
https://huggingface.co/datasets/thirdExec/synthetic-seismic-vlm




