VLABench/vlm_evaluation_v0_test

v0数据集旨在以非交互方式评估VLMs的能力。该数据集的初始版本主要用于帮助用户理解基准测试的结构和设计。数据集按能力维度组织，每个维度有专门的目录，目录下有针对不同任务的子目录。每个任务包含多个示例，每个示例包含原始多视角图像、分割图像、语言指令和真实输出技能序列。

The v0 dataset is designed to evaluate the capabilities of VLMs in a non-interactive manner. This initial version of the dataset is primarily intended to help users understand the structure and design of our benchmark. The dataset is organized by dimensions of ability, with each dimension having a dedicated directory, within which there are multiple subdirectories for task-specific data. Each task includes multiple examples, each of which contains original multi-view images, segmented images for visual prompts, a language instruction, and the ground truth output skill sequence.