Leaf & Gill

项目介绍

Leaf & Gill is an offline plant-and-mushroom identifier that runs entirely in your phone's browser — and it knows when it's guessing. The 609 MB BioCLIP 2 image tower (exported to ONNX fp16, verified cosine 1.00000 against the original) downloads once, after which the app works in airplane mode. Each photo gets a top-5 candidate list with a trust meter, and lookalike species that could kill you are flagged in red. The trust meter itself is built with TabPFN, and every identification the author ran through evaluation and safety tests. Author Mohamed Amine Hamdi wrote in the 2026-10-11 dev.to launch post, verbatim: "I built this with Claude Code as my pair programmer, from the first research question ('can a 609 MB biology model run in iPhone Safari?') to the evaluation and the safety tests." The honest-uncertainty design is what elevates this beyond a demo: instead of confidently misidentifying a death cap as an edible mushroom, the app quantifies its doubt and warns you. For foragers and field biologists that's a safety feature, not a UX nicety. As an engineering artifact, it's also a proof point that a 600 MB vision model is now shippable to a phone browser via WebGPU. MIT, pushed 2026-10-11.