RT-DETR-Edge

22%

by Baidu Research

Real-time DETR optimized for transformer-friendly NPUs. Wins on transformer-class silicon, struggles on legacy NPUs.

Object detectionApache-2.0INT8FP16transformercocodetection

92K downloads 5.4K deploymentsUpdated Apr 2, 2028

Headline:18.6ms · NVIDIA Jetson Orin Nano · FP16

arXiv

Overview Benchmarks6 Sim Results Deploy7 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
N NVIDIA Jetson Thor 2070 TOPS · Module	FP16	6.12	7.99	163 FPS	99.9	102.3	51	Fo’c’sle HIL
N NVIDIA Jetson AGX Orin 275 TOPS · Module	FP16	9.49	11.7	105 FPS	99.8	40.8	51	Fo’c’sle HIL
Q Qualcomm QCS8550 48 TOPS · SoC	INT8	16.1	20.6	62 FPS	98.2	13.1	26	Community
N NVIDIA Jetson Orin Nano 40 TOPS · SoM	INT8	17.3	20.9	58 FPS	98.2	11.4	26	Publisher
A AMD Versal AI Edge VE2302 22 TOPS · SoC	INT8	18.8	23.3	53 FPS	98.9	10.2	26	Community
Apple Neural Engine (M4) 38 TOPS · SoC	FP16	19.6	24.6	51 FPS	99.8	6.8	51	Community
Q Snapdragon 8 Gen 3 NPU 45 TOPS · SoC	Coming Q4 2028
H Hailo-8 26 TOPS · M.2	Not supported
H Hailo-10H 40 TOPS · M.2	Not supported
Q Qualcomm QCS6490 12 TOPS · SoC	Not supported
π Raspberry Pi 5 + Hailo HAT 26 TOPS · HAT	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
G Google Coral Edge TPU 4 TOPS · USB	Not supported
I Intel Movidius Myriad X 4 TOPS · SoC	Not supported
R Rockchip RK3588 6 TOPS · SoC	Not supported

Leader per columnCOCO 2017 val · 1080p input · single-stream, batch=1

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.