ํฌ์ฆ ์ถ์
ํฌ์ฆ ์ถ์ ์ ์ด๋ฏธ์ง์์ ํน์ ์ง์ ์ ์์น๋ฅผ ์๋ณํ๋ ์์
์ผ๋ก, ์ผ๋ฐ์ ์ผ๋ก ํคํฌ์ธํธ๋ผ๊ณ ํฉ๋๋ค. ํคํฌ์ธํธ๋ ๊ด์ , ๋๋๋งํฌ ๋๋ ๊ธฐํ ํน์ง์ ์ธ ํน์ง๊ณผ ๊ฐ์ ๋ฌผ์ฒด์ ๋ค์ํ ๋ถ๋ถ์ ๋ํ๋ผ ์ ์์ต๋๋ค. ํคํฌ์ธํธ์ ์์น๋ ์ผ๋ฐ์ ์ผ๋ก ์ผ๋ จ์ 2D [x, y]
๋๋ 3D [x, y, visible]
์ขํ์
๋๋ค.
ํฌ์ฆ ์ถ์ ๋ชจ๋ธ์ ์ถ๋ ฅ์ ์ผ๋ฐ์ ์ผ๋ก ๊ฐ ํฌ์ธํธ์ ๋ํ ์ ๋ขฐ ์ ์์ ํจ๊ป ์ด๋ฏธ์ง์์ ๊ฐ์ฒด์ ํคํฌ์ธํธ๋ฅผ ๋ํ๋ด๋ ํฌ์ธํธ ์งํฉ์ ๋๋ค. ํฌ์ฆ ์ถ์ ์ ์ฅ๋ฉด์์ ๊ฐ์ฒด์ ํน์ ๋ถ๋ถ๊ณผ ์๋ก์ ๋ํ ์์น๋ฅผ ์๋ณํด์ผ ํ ๋ ์ข์ ์ ํ์ ๋๋ค.
Watch: Pose Estimation with Ultralytics YOLO. |
Watch: ํฌ์ฆ ์ถ์ ์ ์ํ Ultralytics HUB. |
ํ
YOLO11 ํฌ์ฆ ๋ชจ๋ธ์ -pose
์ ๋ฏธ์ฌ, ์ฆ yolo11n-pose.pt
. ์ด ๋ชจ๋ธ์ COCO ํคํฌ์ธํธ ๋ฐ์ดํฐ ์ธํธ๋ฅผ ์ฌ์ฉํ๋ฉฐ ๋ค์ํ ํฌ์ฆ ์ถ์ ์์
์ ์ ํฉํฉ๋๋ค.
In the default YOLO11 pose model, there are 17 keypoints, each representing a different part of the human body. Here is the mapping of each index to its respective body joint:
0: ์ฝ 1: ์ผ์ชฝ ๋ 2: ์ค๋ฅธ์ชฝ ๋ 3: ์ผ์ชฝ ๊ท 4: ์ค๋ฅธ์ชฝ ๊ท 5: ์ผ์ชฝ ์ด๊นจ 6: ์ค๋ฅธ์ชฝ ์ด๊นจ 7: ์ผ์ชฝ ํ๊ฟ์น 8: ์ค๋ฅธ์ชฝ ํ๊ฟ์น 9: ์ผ์ชฝ ์๋ชฉ 10: ์ค๋ฅธ์ชฝ ์๋ชฉ 11: ์ผ์ชฝ ์๋ฉ์ด 12: ์ค๋ฅธ์ชฝ ์๋ฉ์ด 13: ์ผ์ชฝ ๋ฌด๋ฆ 14: ์ค๋ฅธ์ชฝ ๋ฌด๋ฆ 15: ์ผ์ชฝ ๋ฐ๋ชฉ 16: ์ค๋ฅธ์ชฝ ๋ฐ๋ชฉ
๋ชจ๋ธ
YOLO11 pretrained Pose models are shown here. Detect, Segment and Pose models are pretrained on the COCO dataset, while Classify models are pretrained on the ImageNet dataset.
๋ชจ๋ธ์ ์ฒ์ ์ฌ์ฉํ ๋ ์ต์ Ultralytics ๋ฆด๋ฆฌ์ค์์ ์๋์ผ๋ก ๋ค์ด๋ก๋๋ฉ๋๋ค.
๋ชจ๋ธ | ํฌ๊ธฐ (ํฝ์ ) |
mAPpose 50-95 |
mAPpose 50 |
์๋ CPU ONNX (ms) |
Speed T4 TensorRT10 (ms) |
๋งค๊ฐ๋ณ์ (M) |
FLOPs (B) |
---|---|---|---|---|---|---|---|
YOLO11n-pose | 640 | 50.0 | 81.0 | 52.4 ยฑ 0.5 | 1.7 ยฑ 0.0 | 2.9 | 7.6 |
YOLO11s-pose | 640 | 58.9 | 86.3 | 90.5 ยฑ 0.6 | 2.6 ยฑ 0.0 | 9.9 | 23.2 |
YOLO11m-pose | 640 | 64.9 | 89.4 | 187.3 ยฑ 0.8 | 4.9 ยฑ 0.1 | 20.9 | 71.7 |
YOLO11l-pose | 640 | 66.1 | 89.9 | 247.7 ยฑ 1.1 | 6.4 ยฑ 0.1 | 26.2 | 90.7 |
YOLO11x-pose | 640 | 69.5 | 91.1 | 488.0 ยฑ 13.9 | 12.1 ยฑ 0.2 | 58.8 | 203.3 |
- mAPval ๊ฐ์ ๋จ์ผ ๋ชจ๋ธ ๋จ์ผ ์ค์ผ์ผ์ ๋ํ ๊ฒ์
๋๋ค. COCO ํคํฌ์ธํธ val2017 ๋ฐ์ดํฐ ์ธํธ.
๋ณต์ ๋์yolo val pose data=coco-pose.yaml device=0
- ์๋ ๋ฅผ ์ฌ์ฉํ์ฌ COCO ๊ฐ ์ด๋ฏธ์ง์ ๋ํ ํ๊ท ์ ๊ตฌํฉ๋๋ค. Amazon EC2 P4d ์ธ์คํด์ค.
๋ณต์ ๋์yolo val pose data=coco-pose.yaml batch=1 device=0|cpu
๊ธฐ์ฐจ
Train a YOLO11-pose model on the COCO8-pose dataset.
์
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-pose.yaml") # build a new model from YAML
model = YOLO("yolo11n-pose.pt") # load a pretrained model (recommended for training)
model = YOLO("yolo11n-pose.yaml").load("yolo11n-pose.pt") # build from YAML and transfer weights
# Train the model
results = model.train(data="coco8-pose.yaml", epochs=100, imgsz=640)
# Build a new model from YAML and start training from scratch
yolo pose train data=coco8-pose.yaml model=yolo11n-pose.yaml epochs=100 imgsz=640
# Start training from a pretrained *.pt model
yolo pose train data=coco8-pose.yaml model=yolo11n-pose.pt epochs=100 imgsz=640
# Build a new model from YAML, transfer pretrained weights to it and start training
yolo pose train data=coco8-pose.yaml model=yolo11n-pose.yaml pretrained=yolo11n-pose.pt epochs=100 imgsz=640
๋ฐ์ดํฐ ์งํฉ ํ์
YOLO ํฌ์ฆ ๋ฐ์ดํฐ์ ํ์์ ๋ฐ์ดํฐ์ ๊ฐ์ด๋์์ ์์ธํ ํ์ธํ ์ ์์ต๋๋ค. ๊ธฐ์กด ๋ฐ์ดํฐ์ ์ ๋ค๋ฅธ ํ์(์: COCO ๋ฑ)์์ YOLO ํ์์ผ๋ก ๋ณํํ๋ ค๋ฉด JSON2YOLO ๋๊ตฌ( Ultralytics)๋ฅผ ์ฌ์ฉํ์ธ์.
Val
Validate trained YOLO11n-pose model accuracy on the COCO8-pose dataset. No arguments are needed as the model
๊ต์ก ์ ์ง data
๋ฐ ์ธ์๋ฅผ ๋ชจ๋ธ ์์ฑ์ผ๋ก ์ฌ์ฉํฉ๋๋ค.
์
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-pose.pt") # load an official model
model = YOLO("path/to/best.pt") # load a custom model
# Validate the model
metrics = model.val() # no arguments needed, dataset and settings remembered
metrics.box.map # map50-95
metrics.box.map50 # map50
metrics.box.map75 # map75
metrics.box.maps # a list contains map50-95 of each category
์์ธก
Use a trained YOLO11n-pose model to run predictions on images.
์
์ ์ฒด ๋ณด๊ธฐ predict
๋ชจ๋ ์ธ๋ถ ์ ๋ณด์์ ์์ธก ํ์ด์ง๋ก ์ด๋ํฉ๋๋ค.
๋ด๋ณด๋ด๊ธฐ
Export a YOLO11n Pose model to a different format like ONNX, CoreML, etc.
์
Available YOLO11-pose export formats are in the table below. You can export to any format using the format
์ธ์, ์ฆ format='onnx'
๋๋ format='engine'
. ๋ด๋ณด๋ธ ๋ชจ๋ธ์์ ์ง์ ์์ธกํ๊ฑฐ๋ ์ ํจ์ฑ์ ๊ฒ์ฌํ ์ ์์ต๋๋ค. yolo predict model=yolo11n-pose.onnx
. ๋ด๋ณด๋ด๊ธฐ๊ฐ ์๋ฃ๋ ํ ๋ชจ๋ธ์ ๋ํ ์ฌ์ฉ ์๊ฐ ํ์๋ฉ๋๋ค.
ํ์ | format ์ธ์ |
๋ชจ๋ธ | ๋ฉํ๋ฐ์ดํฐ | ์ธ์ |
---|---|---|---|---|
PyTorch | - | yolo11n-pose.pt |
โ | - |
TorchScript | torchscript |
yolo11n-pose.torchscript |
โ | imgsz , optimize , batch |
ONNX | onnx |
yolo11n-pose.onnx |
โ | imgsz , half , dynamic , simplify , opset , batch |
OpenVINO | openvino |
yolo11n-pose_openvino_model/ |
โ | imgsz , half , int8 , batch |
TensorRT | engine |
yolo11n-pose.engine |
โ | imgsz , half , dynamic , simplify , workspace , int8 , batch |
CoreML | coreml |
yolo11n-pose.mlpackage |
โ | imgsz , half , int8 , nms , batch |
TF SavedModel | saved_model |
yolo11n-pose_saved_model/ |
โ | imgsz , keras , int8 , batch |
TF GraphDef | pb |
yolo11n-pose.pb |
โ | imgsz , batch |
TF Lite | tflite |
yolo11n-pose.tflite |
โ | imgsz , half , int8 , batch |
TF Edge TPU | edgetpu |
yolo11n-pose_edgetpu.tflite |
โ | imgsz |
TF.js | tfjs |
yolo11n-pose_web_model/ |
โ | imgsz , half , int8 , batch |
PaddlePaddle | paddle |
yolo11n-pose_paddle_model/ |
โ | imgsz , batch |
NCNN | ncnn |
yolo11n-pose_ncnn_model/ |
โ | imgsz , half , batch |
์ ์ฒด ๋ณด๊ธฐ export
์ธ๋ถ ์ ๋ณด์์ ๋ด๋ณด๋ด๊ธฐ ํ์ด์ง๋ก ์ด๋ํฉ๋๋ค.
์์ฃผ ๋ฌป๋ ์ง๋ฌธ
What is Pose Estimation with Ultralytics YOLO11 and how does it work?
Pose estimation with Ultralytics YOLO11 involves identifying specific points, known as keypoints, in an image. These keypoints typically represent joints or other important features of the object. The output includes the [x, y]
coordinates and confidence scores for each point. YOLO11-pose models are specifically designed for this task and use the -pose
์ ๋ฏธ์ฌ์ ๊ฐ์ yolo11n-pose.pt
. ์ด๋ฌํ ๋ชจ๋ธ์ ๋ค์๊ณผ ๊ฐ์ ๋ฐ์ดํฐ ์ธํธ์ ๋ํด ์ฌ์ ํ์ต๋ฉ๋๋ค. COCO ํคํฌ์ธํธ ๋ค์ํ ํฌ์ฆ ์ถ์ ์์
์ ์ฌ์ฉํ ์ ์์ต๋๋ค. ์์ธํ ๋ด์ฉ์ ํฌ์ฆ ์ถ์ ํ์ด์ง.
How can I train a YOLO11-pose model on a custom dataset?
Training a YOLO11-pose model on a custom dataset involves loading a model, either a new model defined by a YAML file or a pre-trained model. You can then start the training process using your specified dataset and parameters.
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-pose.yaml") # build a new model from YAML
model = YOLO("yolo11n-pose.pt") # load a pretrained model (recommended for training)
# Train the model
results = model.train(data="your-dataset.yaml", epochs=100, imgsz=640)
๊ต์ก์ ๋ํ ์์ธํ ๋ด์ฉ์ ๊ต์ก ์น์ ์ ์ฐธ์กฐํ์ธ์.
How do I validate a trained YOLO11-pose model?
Validation of a YOLO11-pose model involves assessing its accuracy using the same dataset parameters retained during training. Here's an example:
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-pose.pt") # load an official model
model = YOLO("path/to/best.pt") # load a custom model
# Validate the model
metrics = model.val() # no arguments needed, dataset and settings remembered
์์ธํ ๋ด์ฉ์ Val ์น์ ์ ์ฐธ์กฐํ์ธ์.
Can I export a YOLO11-pose model to other formats, and how?
Yes, you can export a YOLO11-pose model to various formats like ONNX, CoreML, TensorRT, and more. This can be done using either Python or the Command Line Interface (CLI).
from ultralytics import YOLO
# Load a model
model = YOLO("yolo11n-pose.pt") # load an official model
model = YOLO("path/to/best.pt") # load a custom trained model
# Export the model
model.export(format="onnx")
์์ธํ ๋ด์ฉ์ ๋ด๋ณด๋ด๊ธฐ ์น์ ์ ์ฐธ์กฐํ์ธ์.
What are the available Ultralytics YOLO11-pose models and their performance metrics?
Ultralytics YOLO11 offers various pretrained pose models such as YOLO11n-pose, YOLO11s-pose, YOLO11m-pose, among others. These models differ in size, accuracy (mAP), and speed. For instance, the YOLO11n-pose model achieves a mAPpose50-95 of 50.4 and an mAPpose50 of 80.1. For a complete list and performance details, visit the Models Section.