spatialverse/SAGE-3D_VLN_Data

SAGE-3D VLN 数据是一个全面的视觉语言导航数据集，包含1000个室内场景中的200万条轨迹-指令对，具有分层指令设计，涵盖从高级语义目标到低级控制命令。该数据集旨在解决现有VLN数据集的局限性，提供因果指令支持和多样化的指令复杂性级别。它包括三个泛化测试分割，并准备好用于动作和图像生成。数据集结构包括测试集和训练集，测试集分为高级VLN、低级VLN和无目标VLN三种任务类型。训练集包含929个场景的轨迹文件，具有丰富的指令多样性。数据集还提供了详细的JSON数据格式和使用说明。

SAGE-3D VLN Data is a comprehensive Vision-Language Navigation dataset featuring 2 million trajectory-instruction pairs across 1,000 indoor scenes, with hierarchical instruction design covering high-level semantic goals to low-level control commands. The dataset addresses limitations in existing VLN datasets by providing causal instruction support and diverse instruction complexity levels. It includes three generalization test splits and is ready for action and image generation. The dataset structure includes test and training sets, with the test set divided into three task types: High-Level VLN, Low-Level VLN, and No-goal VLN. The training set contains trajectory files for 929 scenes with rich instruction diversity. Detailed JSON data format and usage instructions are also provided.