Deployment evaluation of a cucumber size measurement pipeline on Jetson and Raspberry Pi

Manveen Kaur and other authors deployed a curvature-aware cucumber length measurement pipeline (YOLO26n detection, prompt-based segmentation, and mid-arc spline estimation) on NVIDIA Jetson Orin Nano and Raspberry Pi 4, using an RTX 3050 Ti workstation as reference, replaying 50 identical greenhouse footage segments from 7 samples for evaluation. On Jetson, TensorRT FP16 compilation reduced detection latency by 2.57 times, and the same board exhibited two operating points: 72 milliseconds per fruit with 0.84 cm error, or 2.74 seconds per fruit with 0.64 cm error. The authors state that full-image INT8 quantization of the NMS-free detection head fails structurally, returning no detection results at all. When the Raspberry Pi was paired with MobileSAM, the per-fruit computational cost was 2.5 times that of Jetson with SAM, despite 47% lower power consumption, and no thermal throttling occurred under sustained load.

AI 根据报道生成 · 2 days ago更新