MMDuetIT
收藏资源简介:
MMDuetIT数据集用于训练MMDuet模型,并包含用于评估MMDuet的基准。数据集包括密集字幕、时间视频定位和多答案视频问答三个部分。密集字幕部分包含来自Shot2Story和COIN的示例,时间视频定位部分包含来自HiREST、DiDeMo和QueryD的示例,多答案视频问答部分包含自建的Shot2Story-MAGQA-39k数据集。
The MMDuetIT dataset is intended for training the MMDuet model and provides benchmarks for evaluating the MMDuet model. The dataset consists of three components: dense captioning, temporal video grounding, and multi-answer video question answering. The dense captioning component contains examples sourced from Shot2Story and COIN; the temporal video grounding component includes samples from HiREST, DiDeMo and QueryD; and the multi-answer video question answering component features the self-built Shot2Story-MAGQA-39k dataset.
MMDuetIT
数据集描述
MMDuetIT 数据集用于训练 MMDuet 模型,并包含用于评估 MMDuet 的基准数据。数据集的分布如下:
- Dense Captioning
- Shot2Story: 36949 个样本来自 human_anno 子集
- COIN: 4574 个样本来自训练集,视频时长为 2-4 分钟
- Temporal Video Grounding
- HiREST: 459 个样本
- DiDeMo: 24381 个样本
- QueryD: 5392 个样本
- Multi-Answer Grounded Video Question Answering (MAGQA)
- Shot2Story-MAGQA-39k: 包含在数据集中,训练集为
shot2story/annotations/magqa_train-0.25_0.5-earlier.json,测试集为shot2story/annotations/magqa_test.json。问题和答案由 Shot2Story 的人工标注字幕通过 GPT-4o 转换而来。
- Shot2Story-MAGQA-39k: 包含在数据集中,训练集为
相关资源
- 论文: VideoLLM Knows When to Speak: Enhancing Time-Sensitive Video Comprehension with Video-Text Duet Interaction Format
- Github: MMDuet
- 视频演示: YouTube 和 Bilibili
- 模型: MMDuet
引用
如果此工作对你的研究有帮助,请考虑引用:
bibtex @misc{wang2024mmduet, title={VideoLLM Knows When to Speak: Enhancing Time-Sensitive Video Comprehension with Video-Text Duet Interaction Format}, author={Yueqian Wang and Xiaojun Meng and Yuxuan Wang and Jianxin Liang and Jiansheng Wei and Huishuai Zhang and Dongyan Zhao}, year={2024}, eprint={2411.17991}, archivePrefix={arXiv}, primaryClass={cs.CV}, url={https://arxiv.org/abs/2411.17991}, }




