SmellNet: A Large-scale Dataset for Real-world Smell Recognition

SmellNet addresses a critical bottleneck in 'Machine Olfaction': the lack of standardized, large-scale, real-world data. While computer vision and NLP have ImageNet and Common Crawl, olfactory AI has historically relied on small, lab-controlled datasets (like the UCI Gas Sensor Array Drift Dataset). The project's defensibility (4) is driven by the physical difficulty of collecting high-quality chemical sensor data across diverse environments, which is much harder to scrape or generate than text or images. However, with 0 stars and 6 forks at age 4 days, it is currently just a nascent research artifact. Frontier labs (OpenAI/Anthropic) have zero immediate risk here as they lack the hardware-integrated feedback loops for chemical sensing. The primary risk is academic or startup displacement—specifically from companies like Aryballe or Koniku that use proprietary hardware/data loops. The moat is currently 'data gravity' (if the community adopts this as the standard benchmark), but it lacks the community infrastructure to be considered infrastructure-grade yet. It serves as a vital 'Reference Implementation' for researchers trying to bridge the gap between chemical sensors and deep learning.