MoveNet-Lightning

Single-pose lightning model. 30 FPS on a Coral. The pose 'YOLO-Nano'.

PoseApache-2.0INT8posesingle-person

218K downloads 14K deploymentsUpdated Feb 26, 2028

Headline:6.4ms · Google Coral Edge TPU · INT8

Overview Benchmarks8 Sim Results Deploy8 Files Discussion23

Cross-chip benchmark matrix

Every supported chip, in matched-pair runs from the Fo’c’sle HIL lab. Sortable by any column — click a header. Cells where the chip can’t run this model show Not supported.

Chip platform	Quant	Latency p50(ms)	Latency p95(ms)	Throughput	Acc. retention(%)	Power(W)	Memory(MB)	Tested by
Apple Neural Engine (M4) 38 TOPS · SoC	INT8	4.65	6.10	215 FPS	98.2	7.1	5	Publisher
Q Snapdragon 8 Gen 3 NPU 45 TOPS · SoC	INT8	4.72	5.74	212 FPS	98.8	7.1	5	Fo’c’sle HIL
Q Qualcomm QCS8550 48 TOPS · SoC	INT8	5.31	6.60	188 FPS	97.8	11.9	5	Publisher
π Raspberry Pi 5 + Hailo HAT 26 TOPS · HAT	INT8	7.80	10.2	128 FPS	97.8	7.5	5	Publisher
Q Qualcomm QCS6490 12 TOPS · SoC	INT8	10.2	13.6	98 FPS	99.0	8.0	5	Community
R Rockchip RK3588 6 TOPS · SoC	INT8	22.4	27.7	45 FPS	98.8	8.1	5	Community
I Intel Movidius Myriad X 4 TOPS · SoC	INT8	26.7	34.9	37 FPS	98.6	2.3	5	Community
G Google Coral Edge TPU 4 TOPS · USB	INT8	29.7	37.9	34 FPS	97.8	1.4	5	Community
H Hailo-8 26 TOPS · M.2	Not supported
H Hailo-10H 40 TOPS · M.2	Not supported
N NVIDIA Jetson Orin Nano 40 TOPS · SoM	Not supported
N NVIDIA Jetson AGX Orin 275 TOPS · Module	Not supported
N NVIDIA Jetson Thor 2070 TOPS · Module	Not supported
A Ambarella CV5 16 TOPS · SoC	Not supported
A Ambarella CV72 32 TOPS · SoC	Not supported
M MediaTek Genio 700 4 TOPS · SoC	Not supported
A AMD Versal AI Edge VE2302 22 TOPS · SoC	Not supported

Leader per columnCOCO Keypoints val · 640×480 · multi-person

HIL conditions

All numbers measured on Fo’c’sle HIL rigs in Tel Aviv (primary), Munich (secondary), and Pittsburgh (robotics). Single-stream, batch-1, real preprocessing, real downstream consumer. p50/p95 are over 10,000-frame steady-state windows after a 30-second warm-up. Power draw is package power, not wall power. Memory footprint is the resident model + activations footprint at peak — not on-disk.

Submitted publisher numbers are accepted only if they reproduce within ±8% of an HIL-lab matched run on the same chip in the same input mode. Otherwise they live separately under the Discussion tab.