SmellNet-V
收藏资源简介:
SmellNet-V是一个大规模视觉-嗅觉多模态数据集,由韩国科学技术院等机构构建,旨在解决视觉与嗅觉联合表征学习的数据稀缺问题。该数据集基于SmellNet嗅觉数据集,通过语义配对策略将约18万条嗅觉时间序列样本与开放世界的网络图像进行合成配对,数据来源于50种食物和天然成分的便携式气体传感器采集。其创建过程利用了气味在语义类别内的不变性原理,无需昂贵的实地联合采集。该数据集主要应用于多模态感知研究,支持嗅觉分类、跨模态检索及气味源定位等下游任务,以推动人工智能在视觉-嗅觉整合领域的发展。
SmellNet-V is a large-scale visual-olfactory multimodal dataset constructed by institutions including the Korea Advanced Institute of Science and Technology (KAIST), aiming to tackle the data scarcity issue in joint visual and olfactory representation learning. Built upon the original SmellNet olfactory dataset, this dataset adopts a semantic pairing strategy to synthesize paired samples of approximately 180,000 olfactory time-series instances and open-domain web images. The olfactory data was collected via portable gas sensors from 50 categories of foods and natural ingredients. Its development leverages the invariance principle of odors within the same semantic category, eliminating the need for costly on-site joint data collection. This dataset is primarily applied in multimodal perception research, supporting downstream tasks such as olfactory classification, cross-modal retrieval, and odor source localization, so as to advance the progress of artificial intelligence in the field of visual-olfactory integration.



