PolliCount

One phone video, every pollinator visit counted, and each count checkable in the video. THE PROBLEM Pollinator monitoring relies on citizen-science volunteers. In surveys like the UK FIT Count, they watch a flower patch for 10 minutes and note every insect that lands, by eye. Only totals are kept, nothing can be checked, and telling a hoverfly from a bee in a split second is hard. WHAT POLLICOUNT DOES Film the patch with a phone. SAM 3.1 tracks every insect and every flower, a visit counts when an insect lands and stays nearly still for at least 1/3 s, and Muse Spark Open sorts each visitor into its FIT Count group and explains why, with close-ups. Every count links back to its moment in the video: timeline, spotlight, step from visit to visit, slow motion. Counts export to JSON and CSV. Nothing to train, nothing to install. PUSHING THE FRONTIER Filmed from afar, a honeybee is about 30 pixels and SAM 3.1 alone finds no visit at all. PolliCount then finds candidates from their movement and has Muse Spark Open confirm each one: on annotated research videos from Monash University, it now finds 2 visits out of 3, where SAM alone found none. IT WORKS Live app in any browser, phone or computer, with 7 real examples: https://newen-calle--pollicount.us-east.modal.run On hand-checked visitor tracks, Muse Spark Open put every insect in the right group (57/57). The evaluation, a model comparison and the known limits are in the README: measured, not claimed.