spatial_traveres_results_docent-val-Qwen3-VL-8B-Instruct-spatial_no_ref
收藏资源简介:
该数据集是一个用于文本描述生成任务的多模态数据集,特别关注物品场景描述。它包含15个验证集样本,每个样本由多个结构化字段组成,核心字段包括:标注图像(annotated_image)、文本提示词(prompt)、详细的参考描述文本(detailed_reference_description)和生成的描述文本(generated_description)。此外,数据集还提供了物品列表(items)、分段标识(segment_ids)和二维位置坐标(segment_positions),用于细粒度的空间或逻辑分段标注。uuid字段作为唯一标识符,step字段可能表示生成或标注步骤。数据集适用于图像描述生成、视觉语言模型训练与评估、文本生成质量对比等任务,其多字段结构支持对生成描述进行详细分析和评估。
This dataset is a multimodal dataset for text description generation tasks, with a specific focus on object scene descriptions. It contains 15 validation set samples, each composed of multiple structured fields. Core fields include: annotated_image (annotated image), prompt (text prompt), detailed_reference_description (detailed reference description text), and generated_description (generated description text). Additionally, the dataset provides item lists (items), segment identifiers (segment_ids), and 2D position coordinates (segment_positions) for fine-grained spatial or logical segmentation annotation. The uuid field serves as a unique identifier, and the step field may indicate generation or annotation steps. The dataset is suitable for tasks such as image description generation, visual language model training and evaluation, and text generation quality comparison, with its multi-field structure supporting detailed analysis and evaluation of generated descriptions.
- 数据集名称:spatial_traveres_results_docent-val-Qwen3-VL-8B-Instruct-spatial_no_ref
- 数据集地址:https://huggingface.co/datasets/miladalsh/spatial_traveres_results_docent-val-Qwen3-VL-8B-Instruct-spatial_no_ref
- 特征字段:
uuid:字符串类型,唯一标识符。items:字符串列表,数据集条目。segment_ids:整型列表,分段标识符。segment_positions:浮点型列表的列表,分段位置信息。detailed_reference_description:字符串类型,详细参考描述。generated_description:字符串列表,生成的描述。
- 数据划分:仅包含验证集(val),共15个样本,数据集大小为39696字节,下载大小为40439字节。
- 配置文件:默认配置(config_name: default),数据文件路径为
data/val-*。




