YOLOv8-pose
Ultralytics' 2023 YOLO for multi-person pose with COCO's 17 keypoints, n to x, AGPL-3.0 or a paid Enterprise license.
Conditional for size n: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 12.6 ms median, 514 MiB VRAM peak. PyTorch weights, 7 MB.
Conditional for size s: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 12.8 ms median, 584 MiB VRAM peak. PyTorch weights, 22 MB.
Conditional for size m: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 14.0 ms median, 710 MiB VRAM peak. PyTorch weights, 51 MB.
Conditional for size l: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 13.6 ms median, 850 MiB VRAM peak. PyTorch weights, 85 MB.
Conditional for size x: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 13.5 ms median, 1.0 GiB VRAM peak. PyTorch weights, 133 MB.
License
Conditional
Conditional
Conditional
Conditional
Conditional
- Code
- AGPL-3.0-onlychecked 3 Nov 2026
- Weights
- AGPL-3.0-onlyall sizes, checked 3 Nov 2026
AGPL-3.0: shipping it in a product means releasing your source, unless you buy an Ultralytics Enterprise license.
AGPL-3.0: shipping it in a product means releasing your source, unless you buy an Ultralytics Enterprise license.
AGPL-3.0: shipping it in a product means releasing your source, unless you buy an Ultralytics Enterprise license.
AGPL-3.0: shipping it in a product means releasing your source, unless you buy an Ultralytics Enterprise license.
AGPL-3.0: shipping it in a product means releasing your source, unless you buy an Ultralytics Enterprise license.
On our RTX 4090, measured
12.6ms median
p90 12.8 ms, fp16, batch 1, 640x448
- VRAM peak
- 514 MiB
- Status
- Runs as is
Environment, 3 Nov 2026
NVIDIA GeForce RTX 4090, driver 580.173.02, CUDA 13.0, torch 2.14.1+cu130, Python 3.12.9, ultralytics==8.4.189. Baseline 801 MiB. 10 warm-up + 100 timed. Scope: predict(): pre-processing, forward pass and post-processing; input file decoded beforehand.
Accuracy
Not yet scored
Reported50.4 COCO mAPpose 50-95 on COCO val2017 keypointsSource: Ultralytics YOLOv8 docs, pose performance table (mAPpose 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
12.8ms median
p90 13.7 ms, fp16, batch 1, 640x448
- VRAM peak
- 584 MiB
- Status
- Runs as is
Environment, 3 Nov 2026
NVIDIA GeForce RTX 4090, driver 580.173.02, CUDA 13.0, torch 2.14.1+cu130, Python 3.12.9, ultralytics==8.4.189. Baseline 801 MiB. 10 warm-up + 100 timed. Scope: predict(): pre-processing, forward pass and post-processing; input file decoded beforehand.
Accuracy
Not yet scored
Reported60 COCO mAPpose 50-95 on COCO val2017 keypointsSource: Ultralytics YOLOv8 docs, pose performance table (mAPpose 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
14.0ms median
p90 14.4 ms, fp16, batch 1, 640x448
- VRAM peak
- 710 MiB
- Status
- Runs as is
Environment, 3 Nov 2026
NVIDIA GeForce RTX 4090, driver 580.173.02, CUDA 13.0, torch 2.14.1+cu130, Python 3.12.9, ultralytics==8.4.189. Baseline 801 MiB. 10 warm-up + 100 timed. Scope: predict(): pre-processing, forward pass and post-processing; input file decoded beforehand.
Accuracy
Not yet scored
Reported65 COCO mAPpose 50-95 on COCO val2017 keypointsSource: Ultralytics YOLOv8 docs, pose performance table (mAPpose 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
13.6ms median
p90 14.0 ms, fp16, batch 1, 640x448
- VRAM peak
- 850 MiB
- Status
- Runs as is
Environment, 3 Nov 2026
NVIDIA GeForce RTX 4090, driver 580.173.02, CUDA 13.0, torch 2.14.1+cu130, Python 3.12.9, ultralytics==8.4.189. Baseline 801 MiB. 10 warm-up + 100 timed. Scope: predict(): pre-processing, forward pass and post-processing; input file decoded beforehand.
Accuracy
Not yet scored
Reported67.6 COCO mAPpose 50-95 on COCO val2017 keypointsSource: Ultralytics YOLOv8 docs, pose performance table (mAPpose 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
13.5ms median
p90 14.0 ms, fp16, batch 1, 640x448
- VRAM peak
- 1.0 GiB
- Status
- Runs as is
Environment, 3 Nov 2026
NVIDIA GeForce RTX 4090, driver 580.173.02, CUDA 13.0, torch 2.14.1+cu130, Python 3.12.9, ultralytics==8.4.189. Baseline 801 MiB. 10 warm-up + 100 timed. Scope: predict(): pre-processing, forward pass and post-processing; input file decoded beforehand.
Accuracy
Not yet scored
Reported69.2 COCO mAPpose 50-95 on COCO val2017 keypointsSource: Ultralytics YOLOv8 docs, pose performance table (mAPpose 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
Weights
Outputs on the shared test photos
Same photos and one score threshold for every model of this task. Failures are marked.
Sizes
| Size | Latency | VRAM | Accuracy | Weights | Runs | License |
|---|---|---|---|---|---|---|
| n | 12.6 ms640x448 | 514 MiB | Not yet scored50.4 COCO mAPpose 50-95 paper | 7 MB | As is | Conditional |
| s | 12.8 ms640x448 | 584 MiB | Not yet scored60 COCO mAPpose 50-95 paper | 22 MB | As is | Conditional |
| m | 14.0 ms640x448 | 710 MiB | Not yet scored65 COCO mAPpose 50-95 paper | 51 MB | As is | Conditional |
| l | 13.6 ms640x448 | 850 MiB | Not yet scored67.6 COCO mAPpose 50-95 paper | 85 MB | As is | Conditional |
| x | 13.5 ms640x448 | 1.0 GiB | Not yet scored69.2 COCO mAPpose 50-95 paper | 133 MB | As is | Conditional |