CrowdCounter — AI Crowd Count

Point CrowdCounter at a drone photo of a crowd and get a people count in seconds — even when the photo contains thousands of people. HOW IT WORKS CrowdCounter uses an AI density model (DM-Count, trained on the UCF-QNRF benchmark of over a million hand-annotated heads). Instead of trying to draw a box around every person, it predicts a density map of the scene and integrates it into a count — the standard scientific approach for dense crowds where each head is only a few pixels. A color heatmap shows exactly where the people were found, so you can sanity-check every result at a glance. BUILT FOR AERIAL PHOTOS Large drone images are analyzed tile by tile at full resolution, so tiny heads are not lost to downscaling. A 20-megapixel aerial shot works as well as a phone photo. If the picture contains no crowd, the app clearly says so instead of inventing a number. FULLY OFFLINE, PRIVATE BY DESIGN The AI model ships inside the app and runs on your phone. Your photos are never uploaded anywhere — analyze in the field with no signal at all. No account, no ads, no analytics, and no runtime permissions: photos are picked through Android's system photo picker, and camera shots are taken by your camera app. HISTORY THAT SURVIVES Every analyzed photo is stored privately inside the app together with its count and heatmap. Delete the original from your gallery — your analysis history stays complete. Remove entries anytime. ACCURACY On clear daylight crowd photos the typical error is 0-12%, validated against the UCF-QNRF benchmark's human-annotated ground truth. Hard scenes (night shots, extreme density, very small heads) can deviate more. Counts are estimates — a strong starting point, not a certified measurement. FREE AND PREMIUM The free plan includes 2 photo analyses per day. CrowdCounter Premium removes the limit forever with a single one-time purchase — no subscription. Made by Oonak — https://crowdcounter.oonak.ai