TinyEmo-EmoReasonHQ-Claude-1.4k
收藏TinyEmo-EmoReasonHQ-Claude-1.4k
概述
- 任务类别: 文本生成
- 标签: 视觉情感分析、对话、视觉问答、情感推理、多模态LLM、WEBEmo、Claude
- 语言: 英语
- 数据集大小: 1K<n<10K
描述
- 设计目的: 用于训练模型从图像中生成情感推理。
- 数据来源:
- 基于WEBEmo训练数据集(包含213,952张根据Parrott情感分类法标注的图像)。
- 包含1,500个使用Claude 3.5 Sonnet生成的合成高质量情感推理样本。
- 数据集组成: 包含215,452个单任务对话,用于微调模型进行特定的情感推理和视觉情感分析任务。
重要提示
- 该数据集仅包含从Claude Sonnet生成的1.4k高质量推理样本。完整的微调数据集请参见ggcristian/TinyEmo-EmoReason-175k。
示例
- 示例描述: 使用WEBEmo训练数据集中的样本
220_F_49388255_COXFeaC0cUrO4PWyyiKJ8UYH7CkX2TZ4,标注为“nervousness”情感标签,采用指令跟随格式进行微调。 - 示例内容: json { "image": "/root/TFM_CristianGutierrez/data/partitioned_WEBEmo_fine/train/nervousness/220_F_49388255_COXFeaC0cUrO4PWyyiKJ8UYH7CkX2TZ4.jpg", "id": "220_F_49388255_COXFeaC0cUrO4PWyyiKJ8UYH7CkX2TZ4", "conversations": [ { "from": "human", "value": "<image> Provide emotional reasoning about the image, on what you feel and how is the depicted emotion. Also choose the emotion that best corresponds to the image from the following options:
Labels: [nervousness, optimism, sympathy]
Important: note that the correct label is nervousness." }, { "from": "gpt", "value": "Reasoning: The image of a brain with glowing neurons may evoke feelings of nervousness or anxiety due to the association of the glowing neurons with heightened brain activity and potential cognitive overload.
Predicted Label: nervousness" } ] }
引用
@mastersthesis{gutierrez2024tinyemo, title = {TinyEmo: Scaling down Emotional Reasoning via Metric Projection}, author = {Cristian Gutierrez}, year = 2024, month = {September}, address = {Barcelona, Spain}, note = {Available at url{https://ddd.uab.cat/pub/tfg/2024/301610/TFM_CristianGutierrez_2024.pdf}}, school = {Universitat Autonoma de Barcelona (UAB)}, type = {Masters thesis in Computer Vision} }




