YOLOv8-seg
Ultralytics' 2023 YOLO for instance segmentation, five sizes from n to x on COCO's 80 classes, 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: 12.8 ms median, 590 MiB VRAM peak. We scored it 30.1 mask AP on COCO val2017. PyTorch weights, 7 MB.
Conditional for size s: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 13.0 ms median, 652 MiB VRAM peak. We scored it 36.3 mask AP on COCO val2017. PyTorch weights, 23 MB.
Conditional for size m: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 14.4 ms median, 782 MiB VRAM peak. We scored it 40.1 mask AP on COCO val2017. PyTorch weights, 52 MB.
Conditional for size l: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 13.9 ms median, 916 MiB VRAM peak. We scored it 42.1 mask AP on COCO val2017. PyTorch weights, 88 MB.
Conditional for size x: code AGPL-3.0-only, weights AGPL-3.0-only. Measured on our RTX 4090: 14.0 ms median, 1.1 GiB VRAM peak. We scored it 42.8 mask AP on COCO val2017. PyTorch weights, 137 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
12.8ms median
p90 13.4 ms, fp16, batch 1, 640x448
- VRAM peak
- 590 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
30.1mask APmeasured
COCO val2017, mask AP, 5,000 images, fp16, pycocotools==2.0.11 · 2 Nov 2026
Paper 30.5 COCO mAPmask 50-95Ultralytics YOLOv8 docs, segmentation performance table (mAPmask 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
13.0ms median
p90 13.4 ms, fp16, batch 1, 640x448
- VRAM peak
- 652 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
36.3mask APmeasured
COCO val2017, mask AP, 5,000 images, fp16, pycocotools==2.0.11 · 2 Nov 2026
Paper 36.8 COCO mAPmask 50-95Ultralytics YOLOv8 docs, segmentation performance table (mAPmask 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
14.4ms median
p90 15.0 ms, fp16, batch 1, 640x448
- VRAM peak
- 782 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
40.1mask APmeasured
COCO val2017, mask AP, 5,000 images, fp16, pycocotools==2.0.11 · 2 Nov 2026
Paper 40.8 COCO mAPmask 50-95Ultralytics YOLOv8 docs, segmentation performance table (mAPmask 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
13.9ms median
p90 14.2 ms, fp16, batch 1, 640x448
- VRAM peak
- 916 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
42.1mask APmeasured
COCO val2017, mask AP, 5,000 images, fp16, pycocotools==2.0.11 · 2 Nov 2026
Paper 42.6 COCO mAPmask 50-95Ultralytics YOLOv8 docs, segmentation performance table (mAPmask 50-95, docs/en/models/yolov8.md at commit 4fb6485b)
14.0ms median
p90 14.7 ms, fp16, batch 1, 640x448
- VRAM peak
- 1.1 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
42.8mask APmeasured
COCO val2017, mask AP, 5,000 images, fp16, pycocotools==2.0.11 · 2 Nov 2026
Paper 43.4 COCO mAPmask 50-95Ultralytics YOLOv8 docs, segmentation performance table (mAPmask 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 | 12.8 ms640x448 | 590 MiB | 30.1 mask AP, measured30.5 COCO mAPmask 50-95 paper | 7 MB | As is | Conditional |
| s | 13.0 ms640x448 | 652 MiB | 36.3 mask AP, measured36.8 COCO mAPmask 50-95 paper | 23 MB | As is | Conditional |
| m | 14.4 ms640x448 | 782 MiB | 40.1 mask AP, measured40.8 COCO mAPmask 50-95 paper | 52 MB | As is | Conditional |
| l | 13.9 ms640x448 | 916 MiB | 42.1 mask AP, measured42.6 COCO mAPmask 50-95 paper | 88 MB | As is | Conditional |
| x | 14.0 ms640x448 | 1.1 GiB | 42.8 mask AP, measured43.4 COCO mAPmask 50-95 paper | 137 MB | As is | Conditional |