์ฝ˜ํ…์ธ ๋กœ ๊ฑด๋„ˆ๋›ฐ๊ธฐ

๋น ๋ฅธ ์‹œ์ž‘ ๊ฐ€์ด๋“œ: NVIDIA Jetson Ultralytics YOLO11

์ด ํฌ๊ด„์ ์ธ ๊ฐ€์ด๋“œ๋Š” NVIDIA Jetson ์žฅ์น˜์— Ultralytics YOLO11 ๋ฐฐํฌ์— ๋Œ€ํ•œ ์ž์„ธํ•œ ์•ˆ๋‚ด๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ž‘๊ณ  ๊ฐ•๋ ฅํ•œ ์žฅ์น˜์—์„œ YOLO11 ์˜ ๊ธฐ๋Šฅ์„ ์ž…์ฆํ•˜๊ธฐ ์œ„ํ•œ ์„ฑ๋Šฅ ๋ฒค์น˜๋งˆํฌ๋„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค.

New product support

We have updated this guide with the latest NVIDIA Jetson Orin Nano Super Developer Kit which delivers up to 67 TOPS of AI performance โ€” a 1.7X improvement over its predecessor โ€” to seamlessly run the most popular AI models.



Watch: NVIDIA Jetson์„ ์„ค์ •ํ•˜๋Š” ๋ฐฉ๋ฒ• Ultralytics YOLO11

NVIDIA Jetson ์—์ฝ”์‹œ์Šคํ…œ

์ฐธ๊ณ 

This guide has been tested with NVIDIA Jetson Orin Nano Super Developer Kit running the latest stable JetPack release of JP6.1, Seeed Studio reComputer J4012 which is based on NVIDIA Jetson Orin NX 16GB running JetPack release of JP6.0/ JetPack release of JP5.1.3 and Seeed Studio reComputer J1020 v2 which is based on NVIDIA Jetson Nano 4GB running JetPack release of JP4.6.1. It is expected to work across all the NVIDIA Jetson hardware lineup including latest and legacy.

NVIDIA Jetson์ด๋ž€?

NVIDIA Jetson์€ ์—ฃ์ง€ ๋””๋ฐ”์ด์Šค์— ๊ฐ€์†ํ™”๋œ AI(์ธ๊ณต ์ง€๋Šฅ) ์ปดํ“จํŒ…์„ ์ œ๊ณตํ•˜๋„๋ก ์„ค๊ณ„๋œ ์ž„๋ฒ ๋””๋“œ ์ปดํ“จํŒ… ๋ณด๋“œ ์‹œ๋ฆฌ์ฆˆ์ž…๋‹ˆ๋‹ค. ์ด ์ž‘๊ณ  ๊ฐ•๋ ฅํ•œ ์žฅ์น˜๋Š” NVIDIA ์˜ GPU ์•„ํ‚คํ…์ฒ˜๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ๊ตฌ์ถ•๋˜์—ˆ์œผ๋ฉฐ ํด๋ผ์šฐ๋“œ ์ปดํ“จํŒ… ๋ฆฌ์†Œ์Šค์— ์˜์กดํ•  ํ•„์š” ์—†์ด ์žฅ์น˜์—์„œ ์ง์ ‘ ๋ณต์žกํ•œ AI ์•Œ๊ณ ๋ฆฌ์ฆ˜๊ณผ ๋”ฅ ๋Ÿฌ๋‹ ๋ชจ๋ธ์„ ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. Jetson ๋ณด๋“œ๋Š” ๋กœ๋ด‡ ๊ณตํ•™, ์ž์œจ ์ฃผํ–‰ ์ฐจ๋Ÿ‰, ์‚ฐ์—… ์ž๋™ํ™” ๋ฐ ์งง์€ ์ง€์—ฐ ์‹œ๊ฐ„๊ณผ ๋†’์€ ํšจ์œจ์„ฑ์œผ๋กœ ๋กœ์ปฌ์—์„œ AI ์ถ”๋ก ์„ ์ˆ˜ํ–‰ํ•ด์•ผ ํ•˜๋Š” ๊ธฐํƒ€ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์—์„œ ์ž์ฃผ ์‚ฌ์šฉ๋ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ ์ด๋Ÿฌํ•œ ๋ณด๋“œ๋Š” ARM64 ์•„ํ‚คํ…์ฒ˜๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋ฉฐ ๊ธฐ์กด GPU ์ปดํ“จํŒ… ์žฅ์น˜์— ๋น„ํ•ด ์ €์ „๋ ฅ์œผ๋กœ ์‹คํ–‰๋ฉ๋‹ˆ๋‹ค.

NVIDIA Jetson ์‹œ๋ฆฌ์ฆˆ ๋น„๊ต

Jetson Orin์€ ์ด์ „ ์„ธ๋Œ€์— ๋น„ํ•ด ๋Œ€ํญ ํ–ฅ์ƒ๋œ AI ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•˜๋Š” NVIDIA ์•”ํŽ˜์–ด ์•„ํ‚คํ…์ฒ˜๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋Š” NVIDIA Jetson ์ œํ’ˆ๊ตฐ์˜ ์ตœ์‹  ๋ฒ„์ „์ž…๋‹ˆ๋‹ค. ์•„๋ž˜ ํ‘œ๋Š” ์—์ฝ”์‹œ์Šคํ…œ์— ์žˆ๋Š” ๋ช‡ ๊ฐ€์ง€ Jetson ๋””๋ฐ”์ด์Šค๋ฅผ ๋น„๊ตํ•œ ๊ฒƒ์ž…๋‹ˆ๋‹ค.

Jetson AGX Orin 64GB ์ ฏ์Šจ ์˜ค๋ฆฐ NX 16GB Jetson Orin Nano Super Jetson AGX Xavier ์ ฏ์Šจ ์ž๋น„์— NX ์ ฏ์Šจ ๋‚˜๋…ธ
AI ์„ฑ๋Šฅ 275 TOPS 100 TOPS 67 TOPs 32 TOPS 21 TOPS 472 GFLOPS
GPU 2048์ฝ”์–ด NVIDIA ์•”ํŽ˜์–ด ์•„ํ‚คํ…์ฒ˜ GPU (64 Tensor ์ฝ”์–ด) 1024์ฝ”์–ด NVIDIA ์•”ํŽ˜์–ด ์•„ํ‚คํ…์ฒ˜ GPU (32 Tensor ์ฝ”์–ด) 1024์ฝ”์–ด NVIDIA ์•”ํŽ˜์–ด ์•„ํ‚คํ…์ฒ˜ GPU (32 Tensor ์ฝ”์–ด) 512์ฝ”์–ด NVIDIA ๋ณผํƒ€ ์•„ํ‚คํ…์ฒ˜ GPU (64 Tensor ์ฝ”์–ด) 384์ฝ”์–ด NVIDIA Voltaโ„ข ์•„ํ‚คํ…์ฒ˜ GPU , 48๊ฐœ์˜ Tensor ์ฝ”์–ด ํƒ‘์žฌ 128์ฝ”์–ด NVIDIA ๋งฅ์Šค์›ฐโ„ข ์•„ํ‚คํ…์ฒ˜ GPU
GPU ์ตœ๋Œ€ ์ฃผํŒŒ์ˆ˜ 1.3GHz 918MHz 1020 MHz 1377 MHz 1100 MHz 921MHz
CPU 12์ฝ”์–ด NVIDIA Armยฎ Cortex A78AE v8.2 64๋น„ํŠธ CPU 3MB L2 + 6MB L3 8์ฝ”์–ด NVIDIA Armยฎ Cortex A78AE v8.2 64๋น„ํŠธ CPU 2MB L2 + 4MB L3 6์ฝ”์–ด Armยฎ Cortexยฎ-A78AE v8.2 64๋น„ํŠธ CPU 1.5MB L2 + 4MB L3 8์ฝ”์–ด NVIDIA Carmel Armยฎv8.2 64๋น„ํŠธ CPU 8MB L2 + 4MB L3 6์ฝ”์–ด NVIDIA Carmel Armยฎv8.2 64๋น„ํŠธ CPU 6MB L2 + 4MB L3 ์ฟผ๋“œ ์ฝ”์–ด Armยฎ Cortexยฎ-A57 MPCore ํ”„๋กœ์„ธ์„œ
CPU ์ตœ๋Œ€ ์ฃผํŒŒ์ˆ˜ 2.2 GHz 2.0 GHz 1.7 GHz 2.2 GHz 1.9GHz 1.43GHz
๋ฉ”๋ชจ๋ฆฌ 64GB 256๋น„ํŠธ LPDDR5 204.8GB/s 16GB 128๋น„ํŠธ LPDDR5 102.4GB/s 8GB 128-bit LPDDR5 102 GB/s 32GB 256๋น„ํŠธ LPDDR4x 136.5GB/s 8GB 128๋น„ํŠธ LPDDR4x 59.7GB/s 4GB 64๋น„ํŠธ LPDDR4 25.6GB/s"

์ž์„ธํ•œ ๋น„๊ต ํ‘œ๋Š” ๊ณต์‹ NVIDIA Jetson ํŽ˜์ด์ง€์˜ ๊ธฐ์ˆ  ์‚ฌ์–‘ ์„น์…˜์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

NVIDIA ์ œํŠธํŒฉ์ด๋ž€ ๋ฌด์—‡์ธ๊ฐ€์š”?

NVIDIA Jetson ๋ชจ๋“ˆ์„ ๊ตฌ๋™ํ•˜๋Š” JetPack SDK๋Š” ๊ฐ€์žฅ ํฌ๊ด„์ ์ธ ์†”๋ฃจ์…˜์œผ๋กœ, ์—”๋“œํˆฌ์—”๋“œ ๊ฐ€์†ํ™”๋œ AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ๊ตฌ์ถ•์„ ์œ„ํ•œ ์™„๋ฒฝํ•œ ๊ฐœ๋ฐœ ํ™˜๊ฒฝ์„ ์ œ๊ณตํ•˜๊ณ  ์‹œ์žฅ ์ถœ์‹œ ์‹œ๊ฐ„์„ ๋‹จ์ถ•ํ•ฉ๋‹ˆ๋‹ค. ์ ฏํŒฉ์—๋Š” ๋ถ€ํŠธ๋กœ๋”, Linux ์ปค๋„, ์šฐ๋ถ„ํˆฌ ๋ฐ์Šคํฌํ†ฑ ํ™˜๊ฒฝ ๋ฐ GPU ์ปดํ“จํŒ…, ๋ฉ€ํ‹ฐ๋ฏธ๋””์–ด, ๊ทธ๋ž˜ํ”ฝ ๋ฐ ์ปดํ“จํ„ฐ ๋น„์ „ ๊ฐ€์†ํ™”๋ฅผ ์œ„ํ•œ ์ „์ฒด ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ์„ธํŠธ๊ฐ€ ํฌํ•จ๋œ Jetson Linux๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค. ๋˜ํ•œ ํ˜ธ์ŠคํŠธ ์ปดํ“จํ„ฐ์™€ ๊ฐœ๋ฐœ์ž ํ‚คํŠธ ๋ชจ๋‘๋ฅผ ์œ„ํ•œ ์ƒ˜ํ”Œ, ์„ค๋ช…์„œ ๋ฐ ๊ฐœ๋ฐœ์ž ๋„๊ตฌ๊ฐ€ ํฌํ•จ๋˜์–ด ์žˆ์œผ๋ฉฐ ์ŠคํŠธ๋ฆฌ๋ฐ ๋น„๋””์˜ค ๋ถ„์„์„ ์œ„ํ•œ DeepStream, ๋กœ๋ณดํ‹ฑ์Šค๋ฅผ ์œ„ํ•œ Isaac, ๋Œ€ํ™”ํ˜• AI๋ฅผ ์œ„ํ•œ Riva์™€ ๊ฐ™์€ ์ƒ์œ„ ์ˆ˜์ค€์˜ SDK๋ฅผ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค.

ํ”Œ๋ž˜์‹œ ์ œํŠธํŒฉ NVIDIA Jetson

NVIDIA Jetson ์žฅ์น˜๋ฅผ ๊ตฌ์ž…ํ•œ ํ›„ ์ฒซ ๋ฒˆ์งธ ๋‹จ๊ณ„๋Š” NVIDIA JetPack์„ ์žฅ์น˜์— ํ”Œ๋ž˜์‹œํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. NVIDIA Jetson ์žฅ์น˜๋ฅผ ํ”Œ๋ž˜์‹œํ•˜๋Š” ๋ฐฉ๋ฒ•์—๋Š” ์—ฌ๋Ÿฌ ๊ฐ€์ง€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.

  1. ๊ณต์‹ ๊ฐœ๋ฐœ ํ‚คํŠธ( NVIDIA )๋ฅผ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” ๊ฒฝ์šฐ, Jetson Orin Nano ๊ฐœ๋ฐœ์ž ํ‚คํŠธ์™€ ๊ฐ™์€ ์ด๋ฏธ์ง€๋ฅผ ๋‹ค์šด๋กœ๋“œํ•˜๊ณ  ์žฅ์น˜ ๋ถ€ํŒ…์„ ์œ„ํ•ด JetPack์ด ํฌํ•จ๋œ SD ์นด๋“œ๋ฅผ ์ค€๋น„ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  2. ๋‹ค๋ฅธ NVIDIA ๊ฐœ๋ฐœ ํ‚คํŠธ๋ฅผ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” ๊ฒฝ์šฐ SDK ๊ด€๋ฆฌ์ž๋ฅผ ์‚ฌ์šฉํ•˜์—ฌ JetPack์„ ์žฅ์น˜์— ํ”Œ๋ž˜์‹œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  3. ์”จ๋“œ ์ŠคํŠœ๋””์˜ค ์žฌ์ปดํ“จํ„ฐ J4012 ์žฅ์น˜๋ฅผ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” ๊ฒฝ์šฐ JetPack์„ ํฌํ•จ๋œ SSD์— ํ”Œ๋ž˜์‹œํ•  ์ˆ˜ ์žˆ์œผ๋ฉฐ, ์”จ๋“œ ์ŠคํŠœ๋””์˜ค ์žฌ์ปดํ“จํ„ฐ J1020 v2 ์žฅ์น˜๋ฅผ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” ๊ฒฝ์šฐ JetPack์„ eMMC/ SSD์— ํ”Œ๋ž˜์‹œํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
  4. NVIDIA Jetson ๋ชจ๋“ˆ๋กœ ๊ตฌ๋™๋˜๋Š” ๋‹ค๋ฅธ ํƒ€์‚ฌ ์žฅ์น˜๋ฅผ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” ๊ฒฝ์šฐ, ๋ช…๋ น์ค„ ํ”Œ๋ž˜์‹ฑ์„ ๋”ฐ๋ฅด๋Š” ๊ฒƒ์ด ์ข‹์Šต๋‹ˆ๋‹ค.

์ฐธ๊ณ 

์œ„์˜ ๋ฐฉ๋ฒ• 3๊ณผ 4์˜ ๊ฒฝ์šฐ, ์‹œ์Šคํ…œ์„ ํ”Œ๋ž˜์‹œํ•˜๊ณ  ๋””๋ฐ”์ด์Šค๋ฅผ ๋ถ€ํŒ…ํ•œ ํ›„ ๋””๋ฐ”์ด์Šค ํ„ฐ๋ฏธ๋„์—์„œ "sudo apt update && sudo apt install nvidia-jetpack -y"๋ฅผ ์ž…๋ ฅํ•˜์—ฌ ํ•„์š”ํ•œ ๋‚˜๋จธ์ง€ JetPack ๊ตฌ์„ฑ ์š”์†Œ๋ฅผ ๋ชจ๋‘ ์„ค์น˜ํ•˜์„ธ์š”.

์ ฏ์Šจ ๋””๋ฐ”์ด์Šค ๊ธฐ๋ฐ˜์˜ ์ ฏํŒฉ ์ง€์›

์•„๋ž˜ ํ‘œ์—๋Š” ๋‹ค์–‘ํ•œ NVIDIA Jetson ์žฅ์น˜์—์„œ ์ง€์›๋˜๋Š” NVIDIA JetPack ๋ฒ„์ „์ด ๋‚˜์™€ ์žˆ์Šต๋‹ˆ๋‹ค.

์ œํŠธํŒฉ 4 ์ œํŠธํŒฉ 5 ์ œํŠธํŒฉ 6
์ ฏ์Šจ ๋‚˜๋…ธ โœ… โŒ โŒ
Jetson TX2 โœ… โŒ โŒ
์ ฏ์Šจ ์ž๋น„์— NX โœ… โœ… โŒ
Jetson AGX Xavier โœ… โœ… โŒ
Jetson AGX Orin โŒ โœ… โœ…
์ ฏ์Šจ ์˜ค๋ฆฐ NX โŒ โœ… โœ…
์ ฏ์Šจ ์˜ค๋ฆฐ ๋‚˜๋…ธ โŒ โœ… โœ…

Docker๋กœ ๋น ๋ฅด๊ฒŒ ์‹œ์ž‘ํ•˜๊ธฐ

Ultralytics YOLO11 NVIDIA Jetson์„ ์‹œ์ž‘ํ•˜๋Š” ๊ฐ€์žฅ ๋น ๋ฅธ ๋ฐฉ๋ฒ•์€ ๋ฏธ๋ฆฌ ๋นŒ๋“œ๋œ Jetson์šฉ ๋„์ปค ์ด๋ฏธ์ง€๋กœ ์‹คํ–‰ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์œ„์˜ ํ‘œ๋ฅผ ์ฐธ์กฐํ•˜์—ฌ ์†Œ์œ ํ•˜๊ณ  ์žˆ๋Š” Jetson ์žฅ์น˜์— ๋”ฐ๋ผ JetPack ๋ฒ„์ „์„ ์„ ํƒํ•˜์„ธ์š”.

t=ultralytics/ultralytics:latest-jetson-jetpack4
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t
t=ultralytics/ultralytics:latest-jetson-jetpack5
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t
t=ultralytics/ultralytics:latest-jetson-jetpack6
sudo docker pull $t && sudo docker run -it --ipc=host --runtime=nvidia $t

์ด ์ž‘์—…์ด ์™„๋ฃŒ๋˜๋ฉด NVIDIA Jetson ์„น์…˜์˜ TensorRT ์‚ฌ์šฉ์œผ๋กœ ๊ฑด๋„ˆ๋œ๋‹ˆ๋‹ค.

๊ธฐ๋ณธ ์„ค์น˜๋กœ ์‹œ์ž‘ํ•˜๊ธฐ

Docker๋ฅผ ์‚ฌ์šฉํ•˜์ง€ ์•Š๊ณ  ๊ธฐ๋ณธ ์„ค์น˜ํ•˜๋ ค๋ฉด ์•„๋ž˜ ๋‹จ๊ณ„๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.

Run on JetPack 6.1

Ultralytics ํŒจํ‚ค์ง€ ์„ค์น˜

์—ฌ๊ธฐ์„œ๋Š” ์„ ํƒ์  ์ข…์†์„ฑ๊ณผ ํ•จ๊ป˜ Jetson์— Ultralytics ํŒจํ‚ค์ง€๋ฅผ ์„ค์น˜ํ•˜์—ฌ ๋ชจ๋ธ์„ ๋‹ค๋ฅธ ํ˜•์‹์œผ๋กœ ๋‚ด๋ณด๋‚ด๊ธฐ ์œ„ํ•ด PyTorch ๋ชจ๋ธ์„ ๋‹ค๋ฅธ ํ˜•์‹์œผ๋กœ ๋‚ด๋ณด๋‚ผ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. NVIDIA TensorRT ๋‚ด๋ณด๋‚ด๊ธฐ์— ์ฃผ๋กœ ์ดˆ์ ์„ ๋งž์ถœ ๊ฒƒ์ž…๋‹ˆ๋‹ค. TensorRT ์„ ์‚ฌ์šฉํ•˜๋ฉด Jetson ์žฅ์น˜์—์„œ ์ตœ๋Œ€ ์„ฑ๋Šฅ์„ ์–ป์„ ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.

  1. ํŒจํ‚ค์ง€ ๋ชฉ๋ก ์—…๋ฐ์ดํŠธ, pip ์„ค์น˜ ๋ฐ ์ตœ์‹  ๋ฒ„์ „์œผ๋กœ ์—…๊ทธ๋ ˆ์ด๋“œ

    sudo apt update
    sudo apt install python3-pip -y
    pip install -U pip
    
  2. ์„ค์น˜ ultralytics ์„ ํƒ์  ์ข…์†์„ฑ์ด ์žˆ๋Š” pip ํŒจํ‚ค์ง€

    pip install ultralytics[export]
    
  3. ๋””๋ฐ”์ด์Šค ์žฌ๋ถ€ํŒ…

    sudo reboot
    

PyTorch ๋ฐ ํ† ์น˜๋น„์ „ ์„ค์น˜

์œ„์˜ ultralytics ์„ค์น˜๋Š” Torch ๋ฐ Torchvision์„ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ pip๋ฅผ ํ†ตํ•ด ์„ค์น˜๋œ ์ด ๋‘ ํŒจํ‚ค์ง€๋Š” ARM64 ์•„ํ‚คํ…์ฒ˜ ๊ธฐ๋ฐ˜์ธ Jetson ํ”Œ๋žซํผ์—์„œ ์‹คํ–‰ํ•˜๊ธฐ์—๋Š” ํ˜ธํ™˜๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ฏธ๋ฆฌ ๋นŒ๋“œ๋œ PyTorch pip ํœ ์„ ์ˆ˜๋™์œผ๋กœ ์„ค์น˜ํ•˜๊ณ  ์†Œ์Šค์—์„œ Torchvision์„ ์ปดํŒŒ์ผ/์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

์„ค์น˜ torch 2.5.0 ๊ทธ๋ฆฌ๊ณ  torchvision 0.20 according to JP6.1

pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torch-2.5.0a0+872d972e41.nv24.08-cp310-cp310-linux_aarch64.whl
pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torchvision-0.20.0a0+afc54f7-cp310-cp310-linux_aarch64.whl

์ฐธ๊ณ 

๋‹ค๋ฅธ JetPack ๋ฒ„์ „์— ๋Œ€ํ•œ ๋ชจ๋“  ๋‹ค๋ฅธ ๋ฒ„์ „์— ์•ก์„ธ์Šคํ•˜๋ ค๋ฉด PyTorch ์˜ Jetson์šฉ ํŽ˜์ด์ง€ ( PyTorch )๋ฅผ ๋ฐฉ๋ฌธํ•˜์„ธ์š”. ์ž์„ธํ•œ ๋ชฉ๋ก์€ PyTorch, Torchvision ํ˜ธํ™˜์„ฑ ํŽ˜์ด์ง€์—์„œPyTorch ๋ฐ Torchvision ํ˜ธํ™˜์„ฑ ํŽ˜์ด์ง€๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.

์„ค์น˜ cuSPARSELt to fix a dependency issue with torch 2.5.0

wget https://developer.download.nvidia.com/compute/cuda/repos/ubuntu2204/arm64/cuda-keyring_1.1-1_all.deb
sudo dpkg -i cuda-keyring_1.1-1_all.deb
sudo apt-get update
sudo apt-get -y install libcusparselt0 libcusparselt-dev

์„ค์น˜ onnxruntime-gpu

๊ทธ๋ฆฌ๊ณ  onnxruntime-gpu PyPI์—์„œ ํ˜ธ์ŠคํŒ…๋˜๋Š” ํŒจํ‚ค์ง€์—๋Š” aarch64 ๋ฐ”์ด๋„ˆ๋ฆฌ๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด ํŒจํ‚ค์ง€๋ฅผ ์ˆ˜๋™์œผ๋กœ ์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด ํŒจํ‚ค์ง€๋Š” ์ผ๋ถ€ ๋‚ด๋ณด๋‚ด๊ธฐ์— ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.

๋ชจ๋‘ ๋‹ค๋ฅธ onnxruntime-gpu ๋‹ค๋ฅธ JetPack ๋ฐ Python ๋ฒ„์ „์— ํ•ด๋‹นํ•˜๋Š” ํŒจํ‚ค์ง€๊ฐ€ ๋‚˜์—ด๋ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ. ํ•˜์ง€๋งŒ ์—ฌ๊ธฐ์„œ๋Š” ๋‹ค์šด๋กœ๋“œํ•˜์—ฌ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. onnxruntime-gpu 1.20.0 ์™€ ํ•จ๊ป˜ Python3.10 ์ง€์›.

pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/onnxruntime_gpu-1.20.0-cp310-cp310-linux_aarch64.whl

์ฐธ๊ณ 

onnxruntime-gpu ๋ฅผ ์‹คํ–‰ํ•˜๋ฉด numpy ๋ฒ„์ „์ด ์ž๋™์œผ๋กœ ์ตœ์‹  ๋ฒ„์ „์œผ๋กœ ๋˜๋Œ์•„๊ฐ‘๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ numpy๋ฅผ ๋‹ค์‹œ ์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. 1.23.5 ๋ฅผ ์‹คํ–‰ํ•˜์—ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค:

pip install numpy==1.23.5

Run on JetPack 5.1.2

Ultralytics ํŒจํ‚ค์ง€ ์„ค์น˜

์—ฌ๊ธฐ์„œ๋Š” PyTorch ๋ชจ๋ธ์„ ๋‹ค๋ฅธ ๋‹ค๋ฅธ ํ˜•์‹์œผ๋กœ ๋‚ด๋ณด๋‚ผ ์ˆ˜ ์žˆ๋„๋ก ์„ ํƒ์  ์ข…์†์„ฑ๊ณผ ํ•จ๊ป˜ Jetson์— Ultralytics ํŒจํ‚ค์ง€๋ฅผ ์„ค์น˜ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค. NVIDIA TensorRT ๋‚ด๋ณด๋‚ด๊ธฐ์— ์ฃผ๋กœ ์ดˆ์ ์„ ๋งž์ถœ ๊ฒƒ์ž…๋‹ˆ๋‹ค. TensorRT ์„ ์‚ฌ์šฉํ•˜๋ฉด Jetson ์žฅ์น˜์—์„œ ์ตœ๋Œ€ ์„ฑ๋Šฅ์„ ์–ป์„ ์ˆ˜ ์žˆ๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค.

  1. ํŒจํ‚ค์ง€ ๋ชฉ๋ก ์—…๋ฐ์ดํŠธ, pip ์„ค์น˜ ๋ฐ ์ตœ์‹  ๋ฒ„์ „์œผ๋กœ ์—…๊ทธ๋ ˆ์ด๋“œ

    sudo apt update
    sudo apt install python3-pip -y
    pip install -U pip
    
  2. ์„ค์น˜ ultralytics ์„ ํƒ์  ์ข…์†์„ฑ์ด ์žˆ๋Š” pip ํŒจํ‚ค์ง€

    pip install ultralytics[export]
    
  3. ๋””๋ฐ”์ด์Šค ์žฌ๋ถ€ํŒ…

    sudo reboot
    

PyTorch ๋ฐ ํ† ์น˜๋น„์ „ ์„ค์น˜

์œ„์˜ ultralytics ์„ค์น˜๋Š” Torch ๋ฐ Torchvision์„ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ pip๋ฅผ ํ†ตํ•ด ์„ค์น˜๋œ ์ด ๋‘ ํŒจํ‚ค์ง€๋Š” ARM64 ์•„ํ‚คํ…์ฒ˜ ๊ธฐ๋ฐ˜์ธ Jetson ํ”Œ๋žซํผ์—์„œ ์‹คํ–‰ํ•˜๊ธฐ์—๋Š” ํ˜ธํ™˜๋˜์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ฏธ๋ฆฌ ๋นŒ๋“œ๋œ PyTorch pip ํœ ์„ ์ˆ˜๋™์œผ๋กœ ์„ค์น˜ํ•˜๊ณ  ์†Œ์Šค์—์„œ Torchvision์„ ์ปดํŒŒ์ผ/์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

  1. ํ˜„์žฌ ์„ค์น˜๋œ PyTorch ๋ฐ Torchvision ์ œ๊ฑฐ

    pip uninstall torch torchvision
    
  2. ์„ค์น˜ torch 2.1.0 ๊ทธ๋ฆฌ๊ณ  torchvision 0.16.2 according to JP5.1.2

    pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torch-2.1.0a0+41361538.nv23.06-cp38-cp38-linux_aarch64.whl
    pip install https://github.com/ultralytics/assets/releases/download/v0.0.0/torchvision-0.16.2+c6f3977-cp38-cp38-linux_aarch64.whl
    

์ฐธ๊ณ 

๋‹ค๋ฅธ JetPack ๋ฒ„์ „์— ๋Œ€ํ•œ ๋ชจ๋“  ๋‹ค๋ฅธ ๋ฒ„์ „์— ์•ก์„ธ์Šคํ•˜๋ ค๋ฉด PyTorch ์˜ Jetson์šฉ ํŽ˜์ด์ง€ ( PyTorch )๋ฅผ ๋ฐฉ๋ฌธํ•˜์„ธ์š”. ์ž์„ธํ•œ ๋ชฉ๋ก์€ PyTorch, Torchvision ํ˜ธํ™˜์„ฑ ํŽ˜์ด์ง€์—์„œPyTorch ๋ฐ Torchvision ํ˜ธํ™˜์„ฑ ํŽ˜์ด์ง€๋ฅผ ์ฐธ์กฐํ•˜์„ธ์š”.

์„ค์น˜ onnxruntime-gpu

๊ทธ๋ฆฌ๊ณ  onnxruntime-gpu PyPI์—์„œ ํ˜ธ์ŠคํŒ…๋˜๋Š” ํŒจํ‚ค์ง€์—๋Š” aarch64 ๋ฐ”์ด๋„ˆ๋ฆฌ๊ฐ€ ์—†์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ด ํŒจํ‚ค์ง€๋ฅผ ์ˆ˜๋™์œผ๋กœ ์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. ์ด ํŒจํ‚ค์ง€๋Š” ์ผ๋ถ€ ๋‚ด๋ณด๋‚ด๊ธฐ์— ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค.

๋ชจ๋‘ ๋‹ค๋ฅธ onnxruntime-gpu ๋‹ค๋ฅธ JetPack ๋ฐ Python ๋ฒ„์ „์— ํ•ด๋‹นํ•˜๋Š” ํŒจํ‚ค์ง€๊ฐ€ ๋‚˜์—ด๋ฉ๋‹ˆ๋‹ค. ์—ฌ๊ธฐ. ํ•˜์ง€๋งŒ ์—ฌ๊ธฐ์„œ๋Š” ๋‹ค์šด๋กœ๋“œํ•˜์—ฌ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. onnxruntime-gpu 1.17.0 ์™€ ํ•จ๊ป˜ Python3.8 ์ง€์›.

wget https://nvidia.box.com/shared/static/zostg6agm00fb6t5uisw51qi6kpcuwzd.whl -O onnxruntime_gpu-1.17.0-cp38-cp38-linux_aarch64.whl
pip install onnxruntime_gpu-1.17.0-cp38-cp38-linux_aarch64.whl

์ฐธ๊ณ 

onnxruntime-gpu ๋ฅผ ์‹คํ–‰ํ•˜๋ฉด numpy ๋ฒ„์ „์ด ์ž๋™์œผ๋กœ ์ตœ์‹  ๋ฒ„์ „์œผ๋กœ ๋˜๋Œ์•„๊ฐ‘๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ numpy๋ฅผ ๋‹ค์‹œ ์„ค์น˜ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค. 1.23.5 ๋ฅผ ์‹คํ–‰ํ•˜์—ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•ฉ๋‹ˆ๋‹ค:

pip install numpy==1.23.5

NVIDIA Jetson์—์„œ TensorRT ์‚ฌ์šฉ

Ultralytics ์—์„œ ์ง€์›ํ•˜๋Š” ๋ชจ๋“  ๋ชจ๋ธ ๋‚ด๋ณด๋‚ด๊ธฐ ํ˜•์‹ ์ค‘ TensorRT ์€ NVIDIA Jetson ์žฅ์น˜์™€ ํ•จ๊ป˜ ์ž‘์—…ํ•  ๋•Œ ์ตœ๊ณ ์˜ ์ถ”๋ก  ์„ฑ๋Šฅ์„ ์ œ๊ณตํ•˜๋ฉฐ, Jetson๊ณผ ํ•จ๊ป˜ TensorRT ์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด ์ข‹์Šต๋‹ˆ๋‹ค. ์ž์„ธํ•œ ๋‚ด์šฉ์€ TensorRT ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

๋ชจ๋ธ์„ TensorRT ๋กœ ๋ณ€ํ™˜ํ•˜๊ณ  ์ถ”๋ก  ์‹คํ–‰

๋‚ด๋ณด๋‚ธ ๋ชจ๋ธ๋กœ ์ถ”๋ก ์„ ์‹คํ–‰ํ•˜๊ธฐ ์œ„ํ•ด PyTorch ํ˜•์‹์˜ YOLO11n ๋ชจ๋ธ์„ TensorRT ์œผ๋กœ ๋ณ€ํ™˜ํ•ฉ๋‹ˆ๋‹ค.

์˜ˆ

from ultralytics import YOLO

# Load a YOLO11n PyTorch model
model = YOLO("yolo11n.pt")

# Export the model to TensorRT
model.export(format="engine")  # creates 'yolo11n.engine'

# Load the exported TensorRT model
trt_model = YOLO("yolo11n.engine")

# Run inference
results = trt_model("https://ultralytics.com/images/bus.jpg")
# Export a YOLO11n PyTorch model to TensorRT format
yolo export model=yolo11n.pt format=engine  # creates 'yolo11n.engine'

# Run inference with the exported model
yolo predict model=yolo11n.engine source='https://ultralytics.com/images/bus.jpg'

์ฐธ๊ณ 

๋‹ค๋ฅธ ๋ชจ๋ธ ํ˜•์‹์œผ๋กœ ๋ชจ๋ธ์„ ๋‚ด๋ณด๋‚ผ ๋•Œ ์ถ”๊ฐ€ ์ธ์ˆ˜๋ฅผ ์‚ฌ์šฉํ•˜๋ ค๋ฉด ๋‚ด๋ณด๋‚ด๊ธฐ ํŽ˜์ด์ง€๋ฅผ ๋ฐฉ๋ฌธํ•˜์„ธ์š”.

NVIDIA ๋”ฅ๋Ÿฌ๋‹ ์•ก์…€๋Ÿฌ๋ ˆ์ดํ„ฐ(DLA) ์‚ฌ์šฉ

NVIDIA ๋”ฅ ๋Ÿฌ๋‹ ์•ก์…€๋Ÿฌ๋ ˆ์ดํ„ฐ(DLA) ๋Š” ์—๋„ˆ์ง€ ํšจ์œจ๊ณผ ์„ฑ๋Šฅ์„ ์œ„ํ•ด ๋”ฅ ๋Ÿฌ๋‹ ์ถ”๋ก ์„ ์ตœ์ ํ™”ํ•˜๋Š” NVIDIA Jetson ์žฅ์น˜์— ๋‚ด์žฅ๋œ ํŠน์ˆ˜ ํ•˜๋“œ์›จ์–ด ๊ตฌ์„ฑ ์š”์†Œ์ž…๋‹ˆ๋‹ค. GPU ์—์„œ ์ž‘์—…์„ ์˜คํ”„๋กœ๋“œ(๋ณด๋‹ค ์ง‘์ค‘์ ์ธ ํ”„๋กœ์„ธ์Šค๋ฅผ ์œ„ํ•ด ์—ฌ์œ  ๊ณต๊ฐ„์„ ํ™•๋ณด)ํ•จ์œผ๋กœ์จ DLA๋Š” ๋†’์€ ์ฒ˜๋ฆฌ๋Ÿ‰์„ ์œ ์ง€ํ•˜๋ฉด์„œ ๋‚ฎ์€ ์ „๋ ฅ ์†Œ๋น„๋กœ ๋ชจ๋ธ์„ ์‹คํ–‰ํ•  ์ˆ˜ ์žˆ์–ด ์ž„๋ฒ ๋””๋“œ ์‹œ์Šคํ…œ ๋ฐ ์‹ค์‹œ๊ฐ„ AI ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์— ์ด์ƒ์ ์ž…๋‹ˆ๋‹ค.

๋‹ค์Œ Jetson ์žฅ์น˜์—๋Š” DLA ํ•˜๋“œ์›จ์–ด๊ฐ€ ์žฅ์ฐฉ๋˜์–ด ์žˆ์Šต๋‹ˆ๋‹ค:

  • ์ ฏ์Šจ ์˜ค๋ฆฐ NX 16GB
  • ์ ฏ์Šจ AGX ์˜ค๋ฆฐ ์‹œ๋ฆฌ์ฆˆ
  • ์ ฏ์Šจ AGX ์ž๋น„์— ์‹œ๋ฆฌ์ฆˆ
  • ์ ฏ์Šจ ์ž๋น„์— NX ์‹œ๋ฆฌ์ฆˆ

์˜ˆ

from ultralytics import YOLO

# Load a YOLO11n PyTorch model
model = YOLO("yolo11n.pt")

# Export the model to TensorRT with DLA enabled (only works with FP16 or INT8)
model.export(format="engine", device="dla:0", half=True)  # dla:0 or dla:1 corresponds to the DLA cores

# Load the exported TensorRT model
trt_model = YOLO("yolo11n.engine")

# Run inference
results = trt_model("https://ultralytics.com/images/bus.jpg")
# Export a YOLO11n PyTorch model to TensorRT format with DLA enabled (only works with FP16 or INT8)
yolo export model=yolo11n.pt format=engine device="dla:0" half=True  # dla:0 or dla:1 corresponds to the DLA cores

# Run inference with the exported model on the DLA
yolo predict model=yolo11n.engine source='https://ultralytics.com/images/bus.jpg'

์ฐธ๊ณ 

DLA ๋‚ด๋ณด๋‚ด๊ธฐ๋ฅผ ์‚ฌ์šฉํ•  ๋•Œ ์ผ๋ถ€ ๊ณ„์ธต์€ DLA์—์„œ ์‹คํ–‰์ด ์ง€์›๋˜์ง€ ์•Š์„ ์ˆ˜ ์žˆ์œผ๋ฉฐ GPU ๋กœ ํด๋ฐฑ๋˜์–ด ์‹คํ–‰๋ฉ๋‹ˆ๋‹ค. ์ด ํด๋ฐฑ์€ ์ถ”๊ฐ€์ ์ธ ์ง€์—ฐ ์‹œ๊ฐ„์„ ๋ฐœ์ƒ์‹œํ‚ค๊ณ  ์ „๋ฐ˜์ ์ธ ์ถ”๋ก  ์„ฑ๋Šฅ์— ์˜ํ–ฅ์„ ๋ฏธ์น  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค. ๋”ฐ๋ผ์„œ DLA๋Š” GPU ์—์„œ ์ „์ ์œผ๋กœ ์‹คํ–‰๋˜๋Š” TensorRT ์— ๋น„ํ•ด ์ถ”๋ก  ์ง€์—ฐ ์‹œ๊ฐ„์„ ์ค„์ด๊ธฐ ์œ„ํ•œ ๋ชฉ์ ์œผ๋กœ ์„ค๊ณ„๋˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๋Œ€์‹  ์ฒ˜๋ฆฌ๋Ÿ‰์„ ๋Š˜๋ฆฌ๊ณ  ์—๋„ˆ์ง€ ํšจ์œจ์„ฑ์„ ๊ฐœ์„ ํ•˜๋Š” ๊ฒƒ์ด ์ฃผ๋œ ๋ชฉ์ ์ž…๋‹ˆ๋‹ค.

NVIDIA ์ ฏ์Šจ ์˜ค๋ฆฐ YOLO11 ๋ฒค์น˜๋งˆํฌ

YOLO11 benchmarks were run by the Ultralytics team on 10 different model formats measuring speed and accuracy: PyTorch, TorchScript, ONNX, OpenVINO, TensorRT, TF SavedModel, TF GraphDef, TF Lite, PaddlePaddle, NCNN. Benchmarks were run on both NVIDIA Jetson Orin Nano Super Developer Kit and Seeed Studio reComputer J4012 powered by Jetson Orin NX 16GB device at FP32 precision with default input image size of 640.

Comparison Charts

๋ชจ๋“  ๋ชจ๋ธ ๋‚ด๋ณด๋‚ด๊ธฐ๊ฐ€ NVIDIA Jetson์—์„œ ์ž‘๋™ํ•˜์ง€๋งŒ ์•„๋ž˜ ๋น„๊ต ์ฐจํŠธ์—๋Š” PyTorch, TorchScript, TensorRT ๋งŒ ํฌํ•จํ–ˆ๋Š”๋ฐ, ์ด๋Š” Jetson์—์„œ GPU ์„ ์‚ฌ์šฉํ•˜๋ฉฐ ์ตœ์ƒ์˜ ๊ฒฐ๊ณผ๋ฅผ ๋ณด์žฅํ•˜๊ธฐ ๋•Œ๋ฌธ์ž…๋‹ˆ๋‹ค. ๋‹ค๋ฅธ ๋ชจ๋“  ๋‚ด๋ณด๋‚ด๊ธฐ๋Š” CPU ๋งŒ ์‚ฌ์šฉํ•˜๋ฉฐ ์„ฑ๋Šฅ์ด ์œ„์˜ ์„ธ ๊ฐ€์ง€๋ณด๋‹ค ์ข‹์ง€ ์•Š์Šต๋‹ˆ๋‹ค. ์ด ์ฐจํŠธ ๋’ค์˜ ์„น์…˜์—์„œ ๋ชจ๋“  ๋‚ด๋ณด๋‚ด๊ธฐ์— ๋Œ€ํ•œ ๋ฒค์น˜๋งˆํฌ๋ฅผ ์ฐพ์„ ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

NVIDIA Jetson Orin Nano Super Developer Kit

Jetson Orin Nano Super Benchmarks
Benchmarked with Ultralytics 8.3.51

NVIDIA Jetson Orin NX 16GB

Jetson Orin NX 16GB Benchmarks
Benchmarked with Ultralytics 8.3.51

Detailed Comparison Tables

์•„๋ž˜ ํ‘œ๋Š” 10๊ฐ€์ง€ ํ˜•์‹(PyTorch, TorchScript, ONNX, OpenVINO, TensorRT, TF SavedModel , TF GraphDef , TF Lite, PaddlePaddle, NCNN)์— ๋Œ€ํ•œ 5๊ฐ€์ง€ ๋ชจ๋ธ(YOLO11n, YOLO11s, YOLO11m, YOLO11l, YOLO11x)์˜ ๋ฒค์น˜๋งˆํฌ ๊ฒฐ๊ณผ๋กœ, ๊ฐ ์กฐํ•ฉ์˜ ์ƒํƒœ, ํฌ๊ธฐ, mAP50-95(B) ๋ฉ”ํŠธ๋ฆญ ๋ฐ ์ถ”๋ก  ์‹œ๊ฐ„์„ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

NVIDIA Jetson Orin Nano Super Developer Kit

์„ฑ๋Šฅ

ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 5.4 0.6176 21.3
TorchScript โœ… 10.5 0.6100 13.40
ONNX โœ… 10.2 0.6100 7.94
OpenVINO โœ… 10.4 0.6091 57.36
TensorRT (FP32) โœ… 11.9 0.6082 7.60
TensorRT (FP16) โœ… 8.3 0.6096 4.91
TensorRT (INT8) โœ… 5.6 0.3180 3.91
TF SavedModel โœ… 25.8 0.6082 223.98
TF GraphDef โœ… 10.3 0.6082 289.95
TF Lite โœ… 10.3 0.6082 328.29
PaddlePaddle โœ… 20.4 0.6082 530.46
MNN โœ… 10.1 0.6120 74.75
NCNN โœ… 10.2 0.6106 46.12
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 18.4 0.7526 22.00
TorchScript โœ… 36.5 0.7400 21.35
ONNX โœ… 36.3 0.7400 13.91
OpenVINO โœ… 36.4 0.7391 126.95
TensorRT (FP32) โœ… 38.0 0.7400 13.29
TensorRT (FP16) โœ… 21.3 0.7431 7.30
TensorRT (INT8) โœ… 12.2 0.3243 5.25
TF SavedModel โœ… 91.1 0.7400 406.73
TF GraphDef โœ… 36.4 0.7400 629.80
TF Lite โœ… 36.4 0.7400 953.98
PaddlePaddle โœ… 72.5 0.7400 1311.67
MNN โœ… 36.2 0.7392 187.66
NCNN โœ… 36.2 0.7403 122.02
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 38.8 0.7598 33.00
TorchScript โœ… 77.3 0.7643 48.17
ONNX โœ… 76.9 0.7641 29.31
OpenVINO โœ… 77.1 0.7642 313.49
TensorRT (FP32) โœ… 78.7 0.7641 28.21
TensorRT (FP16) โœ… 41.8 0.7653 13.99
TensorRT (INT8) โœ… 23.2 0.4194 9.58
TF SavedModel โœ… 192.7 0.7643 802.30
TF GraphDef โœ… 77.0 0.7643 1335.42
TF Lite โœ… 77.0 0.7643 2842.42
PaddlePaddle โœ… 153.8 0.7643 3644.29
MNN โœ… 76.8 0.7648 503.90
NCNN โœ… 76.8 0.7674 298.78
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 49.0 0.7475 43.00
TorchScript โœ… 97.6 0.7250 62.94
ONNX โœ… 97.0 0.7250 36.33
OpenVINO โœ… 97.3 0.7226 387.72
TensorRT (FP32) โœ… 99.1 0.7250 35.59
TensorRT (FP16) โœ… 52.0 0.7265 17.57
TensorRT (INT8) โœ… 31.0 0.4033 12.37
TF SavedModel โœ… 243.3 0.7250 1116.20
TF GraphDef โœ… 97.2 0.7250 1603.32
TF Lite โœ… 97.2 0.7250 3607.51
PaddlePaddle โœ… 193.9 0.7250 4890.90
MNN โœ… 96.9 0.7222 619.04
NCNN โœ… 96.9 0.7252 352.85
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 109.3 0.8288 81.00
TorchScript โœ… 218.1 0.8308 113.49
ONNX โœ… 217.5 0.8308 75.20
OpenVINO โœ… 217.8 0.8285 508.12
TensorRT (FP32) โœ… 219.5 0.8307 67.32
TensorRT (FP16) โœ… 112.2 0.8248 32.94
TensorRT (INT8) โœ… 61.7 0.4854 20.72
TF SavedModel โœ… 545.0 0.8308 1048.8
TF GraphDef โœ… 217.8 0.8308 2961.8
TF Lite โœ… 217.8 0.8308 7898.8
PaddlePaddle โœ… 434.8 0.8308 9903.68
MNN โœ… 217.3 0.8308 1242.97
NCNN โœ… 217.3 0.8304 850.05

Benchmarked with Ultralytics 8.3.51

NVIDIA Jetson Orin NX 16GB

์„ฑ๋Šฅ

ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 5.4 0.6176 19.50
TorchScript โœ… 10.5 0.6100 13.03
ONNX โœ… 10.2 0.6100 8.44
OpenVINO โœ… 10.4 0.6091 40.83
TensorRT (FP32) โœ… 11.9 0.6100 8.05
TensorRT (FP16) โœ… 8.2 0.6096 4.85
TensorRT (INT8) โœ… 5.5 0.3180 4.37
TF SavedModel โœ… 25.8 0.6082 185.39
TF GraphDef โœ… 10.3 0.6082 244.85
TF Lite โœ… 10.3 0.6082 289.77
PaddlePaddle โœ… 20.4 0.6082 476.52
MNN โœ… 10.1 0.6120 53.37
NCNN โœ… 10.2 0.6106 33.55
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 18.4 0.7526 19.00
TorchScript โœ… 36.5 0.7400 22.90
ONNX โœ… 36.3 0.7400 14.44
OpenVINO โœ… 36.4 0.7391 88.70
TensorRT (FP32) โœ… 37.9 0.7400 14.13
TensorRT (FP16) โœ… 21.6 0.7406 7.55
TensorRT (INT8) โœ… 12.2 0.3243 5.63
TF SavedModel โœ… 91.1 0.7400 317.61
TF GraphDef โœ… 36.4 0.7400 515.99
TF Lite โœ… 36.4 0.7400 838.85
PaddlePaddle โœ… 72.5 0.7400 1170.07
MNN โœ… 36.2 0.7413 125.23
NCNN โœ… 36.2 0.7403 68.13
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 38.8 0.7598 36.50
TorchScript โœ… 77.3 0.7643 52.55
ONNX โœ… 76.9 0.7640 31.16
OpenVINO โœ… 77.1 0.7642 208.57
TensorRT (FP32) โœ… 78.7 0.7640 30.72
TensorRT (FP16) โœ… 41.5 0.7651 14.45
TensorRT (INT8) โœ… 23.3 0.4194 10.19
TF SavedModel โœ… 192.7 0.7643 590.11
TF GraphDef โœ… 77.0 0.7643 998.57
TF Lite โœ… 77.0 0.7643 2486.11
PaddlePaddle โœ… 153.8 0.7643 3236.09
MNN โœ… 76.8 0.7661 335.78
NCNN โœ… 76.8 0.7674 188.43
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 49.0 0.7475 46.6
TorchScript โœ… 97.6 0.7250 66.54
ONNX โœ… 97.0 0.7250 39.55
OpenVINO โœ… 97.3 0.7226 262.44
TensorRT (FP32) โœ… 99.2 0.7250 38.68
TensorRT (FP16) โœ… 51.9 0.7265 18.53
TensorRT (INT8) โœ… 30.9 0.4033 13.36
TF SavedModel โœ… 243.3 0.7250 850.25
TF GraphDef โœ… 97.2 0.7250 1324.60
TF Lite โœ… 97.2 0.7250 3191.24
PaddlePaddle โœ… 193.9 0.7250 4204.97
MNN โœ… 96.9 0.7225 414.41
NCNN โœ… 96.9 0.7252 237.74
ํ˜•์‹ ์ƒํƒœ ๋””์Šคํฌ ํฌ๊ธฐ(MB) mAP50-95(B) ์ถ”๋ก  ์‹œ๊ฐ„(ms/im)
PyTorch โœ… 109.3 0.8288 86.00
TorchScript โœ… 218.1 0.8308 122.43
ONNX โœ… 217.5 0.8307 77.50
OpenVINO โœ… 217.8 0.8285 508.12
TensorRT (FP32) โœ… 219.5 0.8307 76.44
TensorRT (FP16) โœ… 112.0 0.8309 35.99
TensorRT (INT8) โœ… 61.6 0.4854 22.32
TF SavedModel โœ… 545.0 0.8308 1470.06
TF GraphDef โœ… 217.8 0.8308 2549.78
TF Lite โœ… 217.8 0.8308 7025.44
PaddlePaddle โœ… 434.8 0.8308 8364.89
MNN โœ… 217.3 0.8289 827.13
NCNN โœ… 217.3 0.8304 490.29

Benchmarked with Ultralytics 8.3.51

๋‹ค์–‘ํ•œ ๋ฒ„์ „์˜ NVIDIA Jetson ํ•˜๋“œ์›จ์–ด์—์„œ ์‹คํ–‰๋˜๋Š” Seeed Studio๋ฅผ ํ†ตํ•ด ๋” ๋งŽ์€ ๋ฒค์น˜๋งˆํ‚น ๊ฒฐ๊ณผ๋ฅผ ์‚ดํŽด๋ณด์„ธ์š”.

๊ฒฐ๊ณผ ์žฌํ˜„

๋ชจ๋“  ๋‚ด๋ณด๋‚ด๊ธฐ ํ˜•์‹์—์„œ ์œ„์˜ Ultralytics ๋ฒค์น˜๋งˆํฌ๋ฅผ ์žฌํ˜„ํ•˜๋ ค๋ฉด ๋‹ค์Œ ์ฝ”๋“œ๋ฅผ ์‹คํ–‰ํ•˜์„ธ์š”:

์˜ˆ

from ultralytics import YOLO

# Load a YOLO11n PyTorch model
model = YOLO("yolo11n.pt")

# Benchmark YOLO11n speed and accuracy on the COCO8 dataset for all all export formats
results = model.benchmark(data="coco8.yaml", imgsz=640)
# Benchmark YOLO11n speed and accuracy on the COCO8 dataset for all all export formats
yolo benchmark model=yolo11n.pt data=coco8.yaml imgsz=640

๋ฒค์น˜๋งˆํ‚น ๊ฒฐ๊ณผ๋Š” ์‹œ์Šคํ…œ์˜ ์ •ํ™•ํ•œ ํ•˜๋“œ์›จ์–ด ๋ฐ ์†Œํ”„ํŠธ์›จ์–ด ๊ตฌ์„ฑ๊ณผ ๋ฒค์น˜๋งˆํฌ๋ฅผ ์‹คํ–‰ํ•  ๋‹น์‹œ ์‹œ์Šคํ…œ์˜ ํ˜„์žฌ ์ž‘์—…๋Ÿ‰์— ๋”ฐ๋ผ ๋‹ฌ๋ผ์งˆ ์ˆ˜ ์žˆ๋‹ค๋Š” ์ ์— ์œ ์˜ํ•˜์„ธ์š”. ๊ฐ€์žฅ ์‹ ๋ขฐํ•  ์ˆ˜ ์žˆ๋Š” ๊ฒฐ๊ณผ๋ฅผ ์–ป์œผ๋ ค๋ฉด ๋งŽ์€ ์ˆ˜์˜ ์ด๋ฏธ์ง€๊ฐ€ ํฌํ•จ๋œ ๋ฐ์ดํ„ฐ ์„ธํŠธ๋ฅผ ์‚ฌ์šฉํ•˜์„ธ์š”. data='coco8.yaml' (4 val images), ordata='coco.yaml'` (5000๊ฐœ val ์ด๋ฏธ์ง€).

NVIDIA Jetson ์‚ฌ์šฉ ์‹œ ๋ชจ๋ฒ” ์‚ฌ๋ก€

NVIDIA Jetson์„ ์‚ฌ์šฉํ•˜๋Š” ๊ฒฝ์šฐ, NVIDIA Jetson์„ ์‹คํ–‰ํ•˜๋Š” YOLO11 ์—์„œ ์„ฑ๋Šฅ์„ ๊ทน๋Œ€ํ™”ํ•˜๊ธฐ ์œ„ํ•ด ๋”ฐ๋ผ์•ผ ํ•  ๋ช‡ ๊ฐ€์ง€ ๋ชจ๋ฒ” ์‚ฌ๋ก€๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.

  1. ์ตœ๋Œ€ ์ „๋ ฅ ๋ชจ๋“œ ํ™œ์„ฑํ™”

    Jetson์—์„œ ์ตœ๋Œ€ ์ „๋ ฅ ๋ชจ๋“œ๋ฅผ ํ™œ์„ฑํ™”ํ•˜๋ฉด CPU, GPU ์ฝ”์–ด๊ฐ€ ๋ชจ๋‘ ์ผœ์ ธ ์žˆ๋Š”์ง€ ํ™•์ธํ•ฉ๋‹ˆ๋‹ค.

    sudo nvpmodel -m 0
    
  2. Jetson ํด๋ก ์‚ฌ์šฉ

    Jetson ํด๋Ÿญ์„ ํ™œ์„ฑํ™”ํ•˜๋ฉด ๋ชจ๋“  CPU, GPU ์ฝ”์–ด๊ฐ€ ์ตœ๋Œ€ ์ฃผํŒŒ์ˆ˜๋กœ ํด๋Ÿญ๋ฉ๋‹ˆ๋‹ค.

    sudo jetson_clocks
    
  3. Jetson ํ†ต๊ณ„ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜ ์„ค์น˜

    ์ ฏ์Šจ ํ†ต๊ณ„ ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ์‚ฌ์šฉํ•˜์—ฌ ์‹œ์Šคํ…œ ๊ตฌ์„ฑ ์š”์†Œ์˜ ์˜จ๋„๋ฅผ ๋ชจ๋‹ˆํ„ฐ๋งํ•˜๊ณ  CPU, GPU, RAM ์‚ฌ์šฉ๋ฅ , ์ „์› ๋ชจ๋“œ ๋ณ€๊ฒฝ, ์ตœ๋Œ€ ํด๋Ÿญ์œผ๋กœ ์„ค์ •, ์ ฏํŒฉ ์ •๋ณด ํ™•์ธ๊ณผ ๊ฐ™์€ ๊ธฐํƒ€ ์‹œ์Šคํ…œ ์„ธ๋ถ€ ์ •๋ณด๋ฅผ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

    sudo apt update
    sudo pip install jetson-stats
    sudo reboot
    jtop
    

Jetson ํ†ต๊ณ„

๋‹ค์Œ ๋‹จ๊ณ„

NVIDIA Jetson์— YOLO11 ์„ ์„ฑ๊ณต์ ์œผ๋กœ ์„ค์ •ํ•˜์‹  ๊ฒƒ์„ ์ถ•ํ•˜๋“œ๋ฆฝ๋‹ˆ๋‹ค! ์ถ”๊ฐ€ ํ•™์Šต ๋ฐ ์ง€์›์€ Ultralytics YOLO11 ๋ฌธ์„œ์—์„œ ๋” ๋งŽ์€ ๊ฐ€์ด๋“œ๋ฅผ ํ™•์ธํ•˜์„ธ์š”!

์ž์ฃผ ๋ฌป๋Š” ์งˆ๋ฌธ

NVIDIA Jetson ์žฅ์น˜์— Ultralytics YOLO11 ๋ฐฐํฌํ•˜๋ ค๋ฉด ์–ด๋–ป๊ฒŒ ํ•ฉ๋‹ˆ๊นŒ?

NVIDIA Jetson ์žฅ์น˜์— Ultralytics YOLO11 ๋ฐฐํฌํ•˜๋Š” ๊ณผ์ •์€ ๊ฐ„๋‹จํ•ฉ๋‹ˆ๋‹ค. ๋จผ์ € NVIDIA JetPack SDK๋กœ Jetson ์žฅ์น˜๋ฅผ ํ”Œ๋ž˜์‹œํ•ฉ๋‹ˆ๋‹ค. ๊ทธ๋Ÿฐ ๋‹ค์Œ ๋น ๋ฅธ ์„ค์ •์„ ์œ„ํ•ด ์‚ฌ์ „ ๋นŒ๋“œ๋œ Docker ์ด๋ฏธ์ง€๋ฅผ ์‚ฌ์šฉํ•˜๊ฑฐ๋‚˜ ํ•„์š”ํ•œ ํŒจํ‚ค์ง€๋ฅผ ์ˆ˜๋™์œผ๋กœ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค. ๊ฐ ์ ‘๊ทผ ๋ฐฉ์‹์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‹จ๊ณ„๋Š” Docker๋กœ ๋น ๋ฅธ ์‹œ์ž‘ ๋ฐ ๊ธฐ๋ณธ ์„ค์น˜๋กœ ์‹œ์ž‘ ์„น์…˜์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

NVIDIA Jetson ์žฅ์น˜์—์„œ YOLO11 ๋ชจ๋ธ์—์„œ ์–ด๋–ค ์„ฑ๋Šฅ ๋ฒค์น˜๋งˆํฌ๋ฅผ ๊ธฐ๋Œ€ํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๊นŒ?

YOLO11 models have been benchmarked on various NVIDIA Jetson devices showing significant performance improvements. For example, the TensorRT format delivers the best inference performance. The table in the Detailed Comparison Tables section provides a comprehensive view of performance metrics like mAP50-95 and inference time across different model formats.

NVIDIA Jetson์— YOLO11 ๋ฐฐํฌ ์‹œ TensorRT ์„ ์‚ฌ์šฉํ•ด์•ผ ํ•˜๋Š” ์ด์œ ๋Š” ๋ฌด์—‡์ธ๊ฐ€์š”?

TensorRT ์€ ์ตœ์ ์˜ ์„ฑ๋Šฅ์œผ๋กœ ์ธํ•ด NVIDIA Jetson์— YOLO11 ๋ชจ๋ธ์„ ๋ฐฐํฌํ•˜๋Š” ๋ฐ ์ ๊ทน ๊ถŒ์žฅ๋ฉ๋‹ˆ๋‹ค. Jetson์˜ GPU ๊ธฐ๋Šฅ์„ ํ™œ์šฉํ•˜์—ฌ ์ถ”๋ก ์„ ๊ฐ€์†ํ™”ํ•˜์—ฌ ์ตœ๋Œ€์˜ ํšจ์œจ์„ฑ๊ณผ ์†๋„๋ฅผ ๋ณด์žฅํ•ฉ๋‹ˆ๋‹ค. TensorRT ๋กœ ๋ณ€ํ™˜ํ•˜๊ณ  ์ถ”๋ก ์„ ์‹คํ–‰ํ•˜๋Š” ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์ž์„ธํ•œ ๋‚ด์šฉ์€ NVIDIA Jetson์˜ TensorRT ์‚ฌ์šฉ ์„น์…˜์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

NVIDIA Jetson์— PyTorch ๋ฐ Torchvision์„ ์„ค์น˜ํ•˜๋ ค๋ฉด ์–ด๋–ป๊ฒŒ ํ•ด์•ผ ํ•˜๋‚˜์š”?

To install PyTorch and Torchvision on NVIDIA Jetson, first uninstall any existing versions that may have been installed via pip. Then, manually install the compatible PyTorch and Torchvision versions for the Jetson's ARM64 architecture. Detailed instructions for this process are provided in the Installation of PyTorch and Torchvision section.

YOLO11 ์„ ์‚ฌ์šฉํ•  ๋•Œ NVIDIA Jetson์—์„œ ์„ฑ๋Šฅ์„ ๊ทน๋Œ€ํ™”ํ•˜๊ธฐ ์œ„ํ•œ ๋ชจ๋ฒ” ์‚ฌ๋ก€๋Š” ๋ฌด์—‡์ž…๋‹ˆ๊นŒ?

YOLO11 ์„ ์‚ฌ์šฉํ•˜์—ฌ NVIDIA Jetson์˜ ์„ฑ๋Šฅ์„ ๊ทน๋Œ€ํ™”ํ•˜๋ ค๋ฉด ๋‹ค์Œ ๋ชจ๋ฒ” ์‚ฌ๋ก€๋ฅผ ๋”ฐ๋ฅด์‹ญ์‹œ์˜ค:

  1. ์ตœ๋Œ€ ์ „๋ ฅ ๋ชจ๋“œ๋ฅผ ํ™œ์„ฑํ™”ํ•˜์—ฌ CPU ๋ฐ GPU ์ฝ”์–ด๋ฅผ ๋ชจ๋‘ ํ™œ์šฉํ•ฉ๋‹ˆ๋‹ค.
  2. ๋ชจ๋“  ์ฝ”์–ด๋ฅผ ์ตœ๋Œ€ ์ฃผํŒŒ์ˆ˜๋กœ ์‹คํ–‰ํ•˜๋„๋ก Jetson ํด๋Ÿญ์„ ํ™œ์„ฑํ™”ํ•ฉ๋‹ˆ๋‹ค.
  3. ์‹œ์Šคํ…œ ๋ฉ”ํŠธ๋ฆญ์„ ๋ชจ๋‹ˆํ„ฐ๋งํ•˜๊ธฐ ์œ„ํ•ด Jetson Stats ์• ํ”Œ๋ฆฌ์ผ€์ด์…˜์„ ์„ค์น˜ํ•ฉ๋‹ˆ๋‹ค.

๋ช…๋ น์–ด ๋ฐ ์ž์„ธํ•œ ๋‚ด์šฉ์€ NVIDIA Jetson ์‚ฌ์šฉ ์‹œ ๋ชจ๋ฒ” ์‚ฌ๋ก€ ์„น์…˜์„ ์ฐธ์กฐํ•˜์„ธ์š”.

9๊ฐœ์›” ์ „ ์ƒ์„ฑ๋จ โœ๏ธ 10 ์ผ ์ „ ์—…๋ฐ์ดํŠธ๋จ

๋Œ“๊ธ€