FastestDet is a lightweight object detection algorithm designed for resource-constrained environments. It employs an anchor-free approach with a single-scale detector head and cross-grid multiple candidate targets for detection. The model runs on ARM Cortex-A55 CPUs using the NCNN framework, achieving 25.3% mAP@0.5 on COCO at 352x352 resolution with just 0.24M parameters and 70ms inference time on a single core.