CucumberVision at the Edge: Accuracy, Throughput, and Energy of RGB-D Cucumber Sizing on Jetson Orin Nano and Raspberry Pi 4
CucumberVision at the Edge: Accuracy, Throughput, and Energy of RGB-D Cucumber Sizing on Jetson Orin Nano and Raspberry Pi 4
This research evaluates a non-contact fruit sizing pipeline using RGB-D imagery for cucumber length measurement on low-power single-board computers (NVIDIA Jetson Orin Nano and Raspberry Pi 4) compared to an RTX 3050 Ti workstation. The study demonstrates that sub-centimetre accuracy (mean absolute error of 0.58 cm) is preserved across platforms when using TensorRT FP16 compilation on the Jetson. Key findings include detection latency of 72 ms per fruit at 0.84 cm accuracy, energy consumption of 42.2 J per fruit on the Jetson, and structural failure of whole-graph INT8 quantization requiring exclusion of the detection head to restore function. The paper defines an accuracy-throughput-energy envelope with three non-dominated configurations and provides rules for selecting hardware based on specific operational constraints.
The study compares an RGB-D cucumber sizing pipeline on NVIDIA Jetson Orin Nano and Raspberry Pi 4 against an RTX 3050 Ti workstation, reporting sub-centimetre accuracy (MAE 0.58 cm) preserved across platforms with specific latency and energy metrics derived from replayed greenhouse captures.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
Reported numbers
energy_per_fruit
42.2
View original evidence
h INT8 quantization of the NMS-free detection head fails structurally, returning no detections, because one output tensor carries both pixel coordinates and confidences; we show that no calibration strategy can avoid this. Excluding the head restores function at 1.37x. Energy per fruit is 42.2 J on the Jetson, and its 7 W mode consumes 73% more than its 15 W mode. The Raspberry Pi with MobileSAM costs 2.5x more per f
Open source S5detection_latency_084cm
72
View original evidence
rRT FP16 compilation cuts Jetson detection latency 2.57x, giving two operating points on one board: 72 ms per fruit at 0.84 cm, or 2.74 s at 0.64 cm. Four ablations map the design space. For prompted segmentation backends, measurement accuracy is fixed by the backend and reproduces across platforms to within 0.01 cm, while its cost is fixed by the platform and varies 214x, so the two choices are separable. Whole-grap
Open source S4detection_latency_064cm
2.74
View original evidence
rRT FP16 compilation cuts Jetson detection latency 2.57x, giving two operating points on one board: 72 ms per fruit at 0.84 cm, or 2.74 s at 0.64 cm. Four ablations map the design space. For prompted segmentation backends, measurement accuracy is fixed by the backend and reproduces across platforms to within 0.01 cm, while its cost is fixed by the platform and varies 214x, so the two choices are separable. Whole-grap
Open source S4mean_absolute_error
0.58
View original evidence
c spline estimator) on an NVIDIA Jetson Orin Nano and a Raspberry Pi 4, and characterize both against an RTX 3050 Ti workstation by replaying 50 identical greenhouse captures of 7 specimens, with specimen-level cluster-bootstrap statistics. Accuracy is preserved: mean absolute error (MAE) is 0.58 cm (95% CI 0.46-0.69) on both the workstation and the Jetson at matched precision, with a paired difference of zero. Tenso
Open source S3jetson_power_7w_mode
7
View original evidence
h INT8 quantization of the NMS-free detection head fails structurally, returning no detections, because one output tensor carries both pixel coordinates and confidences; we show that no calibration strategy can avoid this. Excluding the head restores function at 1.37x. Energy per fruit is 42.2 J on the Jetson, and its 7 W mode consumes 73% more than its 15 W mode. The Raspberry Pi with MobileSAM costs 2.5x more per f
Open source S5jetson_power_15w_mode
15
View original evidence
h INT8 quantization of the NMS-free detection head fails structurally, returning no detections, because one output tensor carries both pixel coordinates and confidences; we show that no calibration strategy can avoid this. Excluding the head restores function at 1.37x. Energy per fruit is 42.2 J on the Jetson, and its 7 W mode consumes 73% more than its 15 W mode. The Raspberry Pi with MobileSAM costs 2.5x more per f
Open source S5
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
h INT8 quantization of the NMS-free detection head fails structurally, returning no detections, because one output tensor carries both pixel coordinates and confidences; we show that no calibration strategy can avoid this. Excluding the head restores function at 1.37x. Energy per fruit is 42.2 J on the Jetson, and its 7 W mode consumes 73% more than its 15 W mode. The Raspberry Pi with MobileSAM costs 2.5x more per f
Open source S5
rRT FP16 compilation cuts Jetson detection latency 2.57x, giving two operating points on one board: 72 ms per fruit at 0.84 cm, or 2.74 s at 0.64 cm. Four ablations map the design space. For prompted segmentation backends, measurement accuracy is fixed by the backend and reproduces across platforms to within 0.01 cm, while its cost is fixed by the platform and varies 214x, so the two choices are separable. Whole-grap
Open source S4
c spline estimator) on an NVIDIA Jetson Orin Nano and a Raspberry Pi 4, and characterize both against an RTX 3050 Ti workstation by replaying 50 identical greenhouse captures of 7 specimens, with specimen-level cluster-bootstrap statistics. Accuracy is preserved: mean absolute error (MAE) is 0.58 cm (95% CI 0.46-0.69) on both the workstation and the Jetson at matched precision, with a paired difference of zero. Tenso
Open source S3
ruit than the Jetson with SAM despite drawing 47% less power, and it runs without thermal throttling under sustained load. The results define an accuracy-throughput-energy envelope with three non-dominated configurations and give rules for selecting among them.
Open source S6
Source：arXiv Robotics — research abstracts · arxiv.org