Crow Detector — FOMO-style TFLite Trainer

Train a lightweight object detector, quantize it to int8, and export it for the Seeed XIAO ESP32S3 Sense.

Before you start: export your Roboflow dataset in COCO JSON format. You'll get a train folder and a test (or valid) folder, each containing images plus _annotations.coco.json. Click each box below and select the whole folder (not individual files).