Reproduce “New York Smells: A Large Multimodal Dataset for Olfaction” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ege Ozguroglu, Junbang Liang, Ruoshi Liu, Mia Chiquier, Michael DeTienne, Wesley Wei Qian, Alexandra Horowitz, Andrew Owens and 1 more
The "New York Smells" dataset introduces a large-scale collection of 7,000 paired olfactory and visual samples gathered in diverse natural environments, enabling self-supervised learning for general-purpose machine olfaction. Models trained on this multimodal data using raw sensor inputs significantly outperform those using traditional hand-crafted features across various perception tasks, including object recognition and cross-modal retrieval.
Key insights
- Large-scale, multimodal data collected in naturalistic environments is essential for advancing machine olfaction beyond narrow, laboratory-constrained applications.
- Visual information can serve as an effective supervisory signal for self-supervised learning of robust and general-purpose olfactory representations.
- End-to-end processing of raw olfactory sensor signals by neural networks yields significantly richer and more informative representations compared to widely used hand-crafted features like 'smellprints'.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.