Ultralytics çµ±å
Ultralytics ã€ã³ãã°ã¬ãŒã·ã§ã³ã®ããŒãžãžããããïŒãã®ããŒãžã§ã¯ãæ©æ¢°åŠç¿ã®ã¯ãŒã¯ãããŒãåçåããããŒã¿ã»ãã管çã匷åããã¢ãã«ãã¬ãŒãã³ã°ãç°¡çŽ åããå¹ççãªå°å ¥ãä¿é²ããããã«èšèšããããæ§ã ãªããŒã«ããã©ãããã©ãŒã ãšã®ããŒãããŒã·ããã®æŠèŠãæäŸããŸãã
èŠããã ïŒ Ultralytics YOLO11 å±éãšçµ±å
ããŒã¿ã»ããçµ±å
- Roboflow:Ultralytics ã¢ãã«ã®ã·ãŒã ã¬ã¹ãªããŒã¿ã»ãã管çã容æã«ããå ç¢ãªã¢ãããŒã·ã§ã³ãååŠçãæ¡åŒµæ©èœãæäŸã
ãã¬ãŒãã³ã°ã®çµ±å
-
Amazon SageMakerïŒAmazon SageMaker ã掻çšããŠãUltralytics ã¢ãã«ã®æ§ç¯ããã¬ãŒãã³ã°ããããã€ãå¹ççã«è¡ããML ã©ã€ããµã€ã¯ã«ã®ããã®ãªãŒã«ã€ã³ã¯ã³ãã©ãããã©ãŒã ãæäŸããŸãã
-
ClearML:Ultralytics ML ã¯ãŒã¯ãããŒãèªååããå®éšãç£èŠããããŒã ã³ã©ãã¬ãŒã·ã§ã³ãä¿é²ããŸãã
-
Comet ML:Ultralytics ã䜿çšããŠãæ©æ¢°åŠç¿å®éšã远跡ãæ¯èŒãæé©åããããšã§ãã¢ãã«éçºã匷åããŸãã
-
DVC:Ultralytics æ©æ¢°åŠç¿ãããžã§ã¯ãã«ããŒãžã§ã³ç®¡çãå°å ¥ããããŒã¿ãã³ãŒããã¢ãã«ãå¹ççã«åæããŸãã
-
Google ColabïŒGoogle Colab ã䜿çšããŠãã³ã©ãã¬ãŒã·ã§ã³ãšå ±æããµããŒãããã¯ã©ãŠãããŒã¹ã®ç°å¢ã§Ultralytics ã¢ãã«ããã¬ãŒãã³ã°ããã³è©äŸ¡ããŸãã
-
IBM WatsonxïŒIBM Watsonx ãæå 端㮠AI ããŒã«ã容æãªçµ±åãé«åºŠãªã¢ãã«ç®¡çã·ã¹ãã ã«ãããUltralytics ã¢ãã«ã®ãã¬ãŒãã³ã°ãšè©äŸ¡ãã©ã®ããã«ç°¡çŽ åããããã芧ãã ããã
-
JupyterLabïŒJupyterLab ã®ã€ã³ã¿ã©ã¯ãã£ãã§ã«ã¹ã¿ãã€ãºå¯èœãªç°å¢ã䜿çšããŠãUltralytics ã¢ãã«ãç°¡åãã€å¹ççã«ãã¬ãŒãã³ã°ããã³è©äŸ¡ããæ¹æ³ãã芧ãã ããã
-
KaggleïŒããªã€ã³ã¹ããŒã«ãããã©ã€ãã©ãªãGPU ãµããŒããã³ã©ãã¬ãŒã·ã§ã³ãšå ±æã®ããã®æŽ»æ°ããã³ãã¥ããã£ãããã¯ã©ãŠãããŒã¹ã®ç°å¢ã§ãUltralytics ã¢ãã«ã®ãã¬ãŒãã³ã°ãšè©äŸ¡ã«Kaggleãã©ã®ããã«äœ¿çšã§ããããæ¢ããŸãã
-
MLFlow: å®éšãšåçŸæ§ãããããã€ãŸã§ãUltralytics ã¢ãã«ã® ML ã©ã€ããµã€ã¯ã«å šäœãåçåã
-
Neptune:MLOps ã®ããã«èšèšããããã®ã¡ã¿ããŒã¿ã¹ãã¢ã§ãUltralytics ã䜿çšã㊠ML å®éšã®å æ¬çãªãã°ã管çããŸãã
-
Paperspace Gradient Paperspace Gradient ã¯ãã¢ãã«ã®ãã¬ãŒãã³ã°ããã¹ãããããã€ãè¿ éã«è¡ãããã®äœ¿ããããã¯ã©ãŠãããŒã«ãæäŸããããšã§ãYOLO11 ãããžã§ã¯ãã®äœæ¥ãç°¡çŽ åããŸãã
-
Ray TuneïŒUltralytics ã¢ãã«ã®ãã€ããŒãã©ã¡ãŒã¿ãä»»æã®ã¹ã±ãŒã«ã§æé©åããŸãã
-
TensorBoardïŒUltralytics MLã¯ãŒã¯ãããŒãå¯èŠåããã¢ãã«ã¡ããªã¯ã¹ãç£èŠããããŒã ã³ã©ãã¬ãŒã·ã§ã³ãä¿é²ããŸãã
-
Ultralytics HUBïŒäºåã«èšç·ŽãããUltralytics ã¢ãã«ã®ã³ãã¥ããã£ã«ã¢ã¯ã»ã¹ããè²¢ç®ããã
-
Weights & Biases (W&BïŒïŒå®éšãã¢ãã¿ãŒããã¡ããªã¯ã¹ãå¯èŠåããUltralytics ãããžã§ã¯ãã«ãããåçŸæ§ãšã³ã©ãã¬ãŒã·ã§ã³ãä¿é²ããã
-
VS CodeïŒVS Code ã®æ¡åŒµæ©èœã§ãUltralytics ã䜿ã£ãŠéçºã¯ãŒã¯ãããŒãå éããããã®ã³ãŒãã¹ãããããæäŸããŸãããŸããUltralytics ãåŠãã ã䜿ãå§ãããããããã®ãµã³ãã«ãæ¢ããŠãã人ã«ã圹ç«ã¡ãŸãã
-
è£åŒ·ïŒåŒ·åãªç»åè£åŒ·æ©èœã§Ultralytics ã¢ãã«ã匷åããã¢ãã«ã®ããã¹ãæ§ãšæ±åæ§ãåäžãããŸãã
-
SONY IMX500ïŒæé©åãšå±é Ultralytics YOLOv8IMX500 ã»ã³ãµãŒãæèŒãã Raspberry Pi AI ã«ã¡ã©ã®ã¢ãã«ãæé©åããå±éããŸãã
ãããã€ã¡ã³ãã»ã€ã³ãã°ã¬ãŒã·ã§ã³
-
CoreML:CoreMLã¢ããã«ã«ãã£ãŠéçºããããã®ãã¬ãŒã ã¯ãŒã¯ã¯ãiOS ãmacOSãwatchOSãtvOSã®åã¢ããªã±ãŒã·ã§ã³ã«æ©æ¢°åŠç¿ã¢ãã«ãå¹ççã«çµ±åããããã«èšèšããããã®ã§ãã¢ããã«ã®ããŒããŠã§ã¢ã䜿çšããããšã§ãå¹æçãã€å®å šã«ã¢ãã«ãå±éããããšãã§ããŸãã
-
Gradioð NEW: ãªã¢ã«ã¿ã€ã ã®ã€ã³ã¿ã©ã¯ãã£ããªãªããžã§ã¯ãæ€åºãã¢ã®ããã«ãGradio ã§Ultralytics ã¢ãã«ããããã€ããŸãã
-
NCNN:TencentãéçºããNCNN ã¯ãã¢ãã€ã«ããã€ã¹åãã«èª¿æŽãããå¹ççãªãã¥ãŒã©ã«ãããã¯ãŒã¯æšè«ãã¬ãŒã ã¯ãŒã¯ã§ãããAIã¢ãã«ãã¢ããªã«çŽæ¥å°å ¥ã§ããããŸããŸãªã¢ãã€ã«ãã©ãããã©ãŒã ã§ããã©ãŒãã³ã¹ãæé©åã§ããã
-
MNNïŒã¢ãªãããéçºããMNNã¯ãé«å¹çã§è»œéãªãã£ãŒãã©ãŒãã³ã°ãã¬ãŒã ã¯ãŒã¯ã§ããããã£ãŒãã©ãŒãã³ã°ã¢ãã«ã®æšè«ãšåŠç¿ããµããŒãããããã€ã¹äžã§ã®æšè«ãšåŠç¿ã«ãããŠæ¥çããªãŒãããããã©ãŒãã³ã¹ãçºæ®ããã
-
Neural Magic:Quantization Aware Training (QAT)ãšãã«ãŒãã³ã°æè¡ã掻çšããUltralytics ã®ã¢ãã«ãæé©åããããšã§ãåªããããã©ãŒãã³ã¹ãšãµã€ãºã®çž®å°ãå®çŸã
-
ONNX:ã«ãã£ãŠäœæããããªãŒãã³ãœãŒã¹ãã©ãŒãããã Microsoftæ§ã ãªãã¬ãŒã ã¯ãŒã¯éã§ã®AIã¢ãã«ã®è»¢éã容æã«ããUltralytics ã¢ãã«ã®æ±çšæ§ãšå±éã®æè»æ§ã匷åããã
-
OpenVINOIntel Intel CPU æ§ã ãªãã©ãããã©ãŒã ã§GPU ã³ã³ãã¥ãŒã¿ããžã§ã³ã¢ãã«ãå¹ççã«æé©åããå±éããããã®ããŒã«ãããã
-
PaddlePaddle:ãã€ãã¥ã«ãããªãŒãã³ãœãŒã¹ã®ãã£ãŒãã©ãŒãã³ã°ãã©ãããã©ãŒã ãPaddlePaddle ã¯ãAIã¢ãã«ã®å¹ççãªå±éãå¯èœã«ããç£æ¥ã¢ããªã±ãŒã·ã§ã³ã®ã¹ã±ãŒã©ããªãã£ã«çŠç¹ãåœãŠãŠããã
-
TF GraphDef:ã«ãã£ãŠéçºãããã GoogleGraphDef ã¯èšç®ã°ã©ããè¡šçŸããããã®TensorFlow ã®ãã©ãŒãããã§ãããå€æ§ãªããŒããŠã§ã¢éã§ã®æ©æ¢°åŠç¿ã¢ãã«ã®æé©åãããå®è¡ãå¯èœã«ããã
-
TF SavedModel:éçºè GoogleTF SavedModel ã¯ãã¢ãã«ã®ãŠãããŒãµã«ã»ã·ãªã¢ã©ã€ãŒãŒã·ã§ã³ã»ãã©ãŒãããã§ãã TensorFlowãµãŒããŒãããšããžããã€ã¹ãŸã§ãå¹ åºããã©ãããã©ãŒã ã§ã¢ãã«ãç°¡åã«å ±æããå±éããããšãã§ããŸãã
-
TF.jsïŒã«ãã£ãŠéçºãããã Googleã«ãã£ãŠéçºããããã©ãŠã¶ãš Node.js ã§ã®æ©æ¢°åŠç¿ã容æã«ãããTF.js ã¯ãJavaScript ããŒã¹ã® ML ã¢ãã«ã®ãããã€ãå¯èœã«ããã
-
TFLiteéçºå GoogleTFLiteã¯ãã¢ãã€ã«ããã€ã¹ããšããžããã€ã¹äžã§æ©æ¢°åŠç¿ã¢ãã«ãå±éããããã®è»œéãã¬ãŒã ã¯ãŒã¯ã§ãæå°éã®ã¡ã¢ãªãããããªã³ãã§é«éãã€å¹ççãªæšè«ãå®çŸããŸãã
-
TFLite EdgeTPU: TFLiteã«ãã£ãŠéçºãããŸããã GoogleEdge TPUäžã§TensorFlow Liteã¢ãã«ãæé©åããããã«éçºããããã®ã¢ãã«ãã©ãŒãããã¯ãé«éã§å¹ççãªãšããžã³ã³ãã¥ãŒãã£ã³ã°ãä¿èšŒããŸãã
-
TensorRT:ã«ãã£ãŠéçºãããŸããã NVIDIAãã®é«æ§èœãã£ãŒãã©ãŒãã³ã°æšè«ãã¬ãŒã ã¯ãŒã¯ãšã¢ãã«ãã©ãŒãããã¯ãNVIDIA GPUäžã§AIã¢ãã«ãé«éãã€å¹ççã«æé©åããåççãªå°å ¥ãå®çŸããŸãã
-
TorchScript:ã®äžéšãšããŠéçºãããã PyTorchãã¬ãŒã ã¯ãŒã¯ã®äžéšãšããŠéçºãããTorchScript ã¯ãPython ã«äŸåããããšãªããæ§ã ãªæ¬çªç°å¢ã«ãããŠæ©æ¢°åŠç¿ã¢ãã«ã®å¹ççãªå®è¡ãšãããã€ãå¯èœã«ããã
茞åºãã©ãŒããã
ãŸããããŸããŸãªç°å¢ã§ã®å±éã®ããã«ãããŸããŸãªã¢ãã«ã®ãšã¯ã¹ããŒã圢åŒããµããŒãããŠããŸããå©çšå¯èœãªãã©ãŒãããã¯ä»¥äžã®éãã§ãïŒ
ãã©ãŒããã | format è°è« |
ã¢ãã« | ã¡ã¿ããŒã¿ | è°è« |
---|---|---|---|---|
PyTorch | - | yolo11n.pt |
â | - |
TorchScript | torchscript |
yolo11n.torchscript |
â | imgsz , optimize , batch |
ONNX | onnx |
yolo11n.onnx |
â | imgsz , half , dynamic , simplify , opset , batch |
OpenVINO | openvino |
yolo11n_openvino_model/ |
â | imgsz , half , int8 , batch |
TensorRT | engine |
yolo11n.engine |
â | imgsz , half , dynamic , simplify , workspace , int8 , batch |
CoreML | coreml |
yolo11n.mlpackage |
â | imgsz , half , int8 , nms , batch |
TF SavedModel | saved_model |
yolo11n_saved_model/ |
â | imgsz , keras , int8 , batch |
TF GraphDef | pb |
yolo11n.pb |
â | imgsz , batch |
TF ã©ã€ã | tflite |
yolo11n.tflite |
â | imgsz , half , int8 , batch |
TF ãšããžTPU | edgetpu |
yolo11n_edgetpu.tflite |
â | imgsz |
TF.js | tfjs |
yolo11n_web_model/ |
â | imgsz , half , int8 , batch |
PaddlePaddle | paddle |
yolo11n_paddle_model/ |
â | imgsz , batch |
MNN | mnn |
yolo11n.mnn |
â | imgsz , batch , int8 , half |
NCNN | ncnn |
yolo11n_ncnn_model/ |
â | imgsz , half , batch |
IMX500 | imx |
yolov8n_imx_model/ |
â | imgsz , int8 |
åçµ±åã®è©³çŽ°ãšãUltralytics ãæ倧éã«æŽ»çšããæ¹æ³ã«ã€ããŠã¯ããªã³ã¯ãã芧ãã ãããå
šæãèŠã export
è©³çŽ°ã¯ èŒžåº ããŒãžãåç
§ããããã
çµ±åãžã®è²¢ç®
ç§ãã¡ã¯ãã³ãã¥ããã£ãUltralytics YOLO ãä»ã®ãã¯ãããžãŒãããŒã«ããã©ãããã©ãŒã ãšã©ã®ããã«çµ±åããŠããããèŠãããšã«ãã€ãè奮ããŠããŸãïŒYOLO ãšæ°ããã·ã¹ãã ãšã®çµ±åã«æåããå ŽåããŸãã¯å ±æãã貎éãªæŽå¯ããæã¡ã®å Žåã¯ãçµ±åããã¥ã¡ã³ããžã®è²¢ç®ããæ€èšãã ããã
ã¬ã€ããŸãã¯ãã¥ãŒããªã¢ã«ãæžãããšã§ãç§ãã¡ã®ããã¥ã¡ã³ããæ¡åŒµããã³ãã¥ããã£ã®ããã«ãªãå®äŸãæäŸããããšãã§ããŸããããã¯ãUltralytics YOLO ãåãå·»ãæé·äžã®ãšã³ã·ã¹ãã ã«è²¢ç®ããåªããæ¹æ³ã§ãã
å¯çš¿ããã«ã¯ãå¯çš¿ã¬ã€ãã§ãã«ãªã¯ãšã¹ã (PR) ð ïž ã®æåºæ¹æ³ãã確èªãã ãããçæ§ã®ãæçš¿ããåŸ ã¡ããŠãããŸãïŒ
Ultralytics YOLO ãšã³ã·ã¹ãã ãããåºç¯ã§æ©èœè±å¯ãªãã®ã«ããããã«ååããã ðïŒ
ããããã質å
Ultralytics HUBãšã¯äœããMLã®ã¯ãŒã¯ãããŒãã©ã®ããã«å¹çåããã®ãïŒ
Ultralytics HUBã¯ãUltralytics ã¢ãã«ã®æ©æ¢°åŠç¿ïŒMLïŒã¯ãŒã¯ãããŒãã·ãŒã ã¬ã¹ãã€å¹ççã«ããããã«èšèšãããã¯ã©ãŠãããŒã¹ã®ãã©ãããã©ãŒã ã§ãããã®ããŒã«ã䜿çšããããšã§ãããŒã¿ã»ããã®ã¢ããããŒããã¢ãã«ã®ãã¬ãŒãã³ã°ããªã¢ã«ã¿ã€ã ãã©ããã³ã°ã®å®è¡ãYOLO11 ã¢ãã«ã®ãããã€ããè±å¯ãªã³ãŒãã£ã³ã°ã¹ãã«ãå¿ èŠãšããã«ç°¡åã«è¡ãããšãã§ããŸããUltralytics HUBã®ããŒãžã§äž»ãªæ©èœã調ã¹ãã¯ã€ãã¯ã¹ã¿ãŒãã¬ã€ãã§ããã«å§ããããšãã§ããŸãã
Ultralytics YOLO ã¢ãã«ãšRoboflow ãçµ±åããŠããŒã¿ã»ããã管çããã«ã¯ïŒ
Ultralytics YOLO ã¢ãã«ãRoboflow ãšçµ±åããããšã§ãã¢ãããŒã·ã§ã³ãååŠçãå¢åŒ·ã®ããã®å ç¢ãªããŒã«ãæäŸããããŒã¿ã»ããã®ç®¡çã匷åããŸããéå§ããã«ã¯ Roboflowçµ±åããŒãžã®ã¹ãããã«åŸã£ãŠãã ããããã®ããŒãããŒã·ããã«ãããæ£ç¢ºã§ããã¹ããªYOLO ã¢ãã«ã®éçºã«äžå¯æ¬ ãªãå¹ççãªããŒã¿ã»ããã®åãæ±ããä¿èšŒãããã
MLFlowã䜿çšããŠãUltralytics ã¢ãã«ã®ããã©ãŒãã³ã¹ã远跡ã§ããŸããïŒ
ã¯ããã§ããŸããMLFlow ãUltralytics ã¢ãã«ãšçµ±åããããšã§ãå®éšã远跡ããåçŸæ§ãåäžãããML ã©ã€ããµã€ã¯ã«å šäœãåçåããããšãã§ããŸãããã®çµ±åãèšå®ããããã®è©³çŽ°ãªæé ã¯ãMLFlowã®çµ±åããŒãžã«ãããŸãããã®çµ±åã¯ãã¢ãã«ã¡ããªã¯ã¹ã®ã¢ãã¿ãªã³ã°ãMLã¯ãŒã¯ãããŒã®å¹ççãªç®¡çã«ç¹ã«åœ¹ç«ã¡ãŸãã
YOLO11 ã¢ãã«ã®æé©åã«Neural Magic ã䜿çšããã¡ãªããã¯ïŒ
Neural Magic Quantization Aware Training (QAT)ããã«ãŒãã³ã°ã®ãããªãã¯ããã¯ã掻çšããããšã§ãYOLO11 ã®ã¢ãã«ãæé©åãããªãœãŒã¹ãéãããããŒããŠã§ã¢ã§ããåªããããã©ãŒãã³ã¹ãçºæ®ãããé«å¹çã§ããå°ããªã¢ãã«ãå®çŸããŸããçµ±åããŒãž Neural Magicã€ã³ãã°ã¬ãŒã·ã§ã³ã»ããŒãžãã芧ãã ãããããã¯ç¹ã«ãšããžããã€ã¹ã§ã®å±éã«æçã§ãã
ã€ã³ã¿ã©ã¯ãã£ããªãã¢ã®ããã«ãGradioã§Ultralytics YOLO ã¢ãã«ããããã€ããã«ã¯ïŒ
ã€ã³ã¿ã©ã¯ãã£ããªç©äœæ€åºãã¢ã®ããã«ãUltralytics YOLO ã¢ãã«ã Gradio ã§ãããã€ããã«ã¯ãGradioçµ±åããŒãžã«æŠèª¬ãããŠããæé ã«åŸãããšãã§ããŸããGradioã¯ããªã¢ã«ã¿ã€ã ã®ã¢ãã«æšè«ã®ããã®äœ¿ãããããŠã§ãã€ã³ã¿ãŒãã§ãŒã¹ãäœæããããšãã§ããéçºè ãšãšã³ããŠãŒã¶ãŒã®äž¡æ¹ã«é©ãããŠãŒã¶ãŒãã¬ã³ããªãŒãªåœ¢åŒã§ãYOLO ã¢ãã«ã®æ©èœã玹ä»ããããã®åªããããŒã«ãšãªã£ãŠããŸãã
ãããã®äžè¬çãªè³ªåã«å¯Ÿå¿ããããšã§ããŠãŒã¶ãŒãšã¯ã¹ããªãšã³ã¹ãåäžãããUltralytics 補åã®åŒ·åãªæ©èœã«é¢ãã貎éãªæŽå¯ãæäŸããããšãç®çãšããŠããŸãããããã®FAQãåãå ¥ããããšã§ãããã¥ã¡ã³ããå å®ãããã ãã§ãªããUltralytics ãŠã§ããµã€ããžã®ãªãŒã¬ããã¯ãªãã©ãã£ãã¯ãå¢ããããšãã§ããŸãã