泚éä»ãã³ã³ãã¥ãŒã¿ããžã§ã³ããŒã¿ã®ããŒã¿ååŠçææ³
ã¯ããã«
ã³ã³ãã¥ãŒã¿ããžã§ã³ãããžã§ã¯ãã®ç®æšãå®ããããŒã¿ãåéãã泚éãä»ãããã次ã®ã¹ãããã¯æ³šéä»ãããŒã¿ãååŠçããã¢ãã«ãã¬ãŒãã³ã°ã®æºåãããããšã§ããã¯ãªãŒã³ã§äžè²«æ§ã®ããããŒã¿ã¯ãåªããããã©ãŒãã³ã¹ãçºæ®ããã¢ãã«ãäœæããããã«äžå¯æ¬ ã§ãã
ååŠçã¯ãç»åã®ãµã€ãºå€æŽããã¯ã»ã«å€ã®æ£èŠåãããŒã¿ã»ããã®å¢åŒ·ãããŒã¿ããã¬ãŒãã³ã°ã»ãããæ€èšŒã»ããããã¹ãã»ããã«åå²ããããšãå«ããã³ã³ãã¥ãŒã¿ããžã§ã³ãããžã§ã¯ãã¯ãŒã¯ãããŒã®ã¹ãããã§ããããŒã¿ã®ã¯ãªãŒãã³ã°ã«äžå¯æ¬ ãªãã¯ããã¯ãšãã¹ããã©ã¯ãã£ã¹ãæ¢ã£ãŠã¿ãŸãããïŒ
ããŒã¿ã®ååŠçã®éèŠæ§
ç§ãã¡ã¯ãã§ã«ãè€æ°ã®èæ ®äºé ã念é ã«çœ®ããŠæ éã«ããŒã¿ãåéãã泚éãä»ããŠããŸããã§ã¯ãã³ã³ãã¥ãŒã¿ãŒããžã§ã³ãããžã§ã¯ãã«ãšã£ãŠããŒã¿ã®ååŠçãããã»ã©éèŠãªçç±ã¯äœã§ãããããããŒã¿ã®ååŠçãšã¯ãããŒã¿ããã¬ãŒãã³ã°ã«é©ãã圢åŒã«ããŠãèšç®è² è·ã軜æžããã¢ãã«ã®ããã©ãŒãã³ã¹ãåäžãããããšã§ããããã§ã¯ãååŠçã§å¯ŸåŠããçããŒã¿ã®äžè¬çãªåé¡ãããã€ã玹ä»ããŸãã
- NoiseïŒãã€ãºïŒïŒããŒã¿ã«å«ãŸããç¡é¢ä¿ãŸãã¯ã©ã³ãã ãªå€åã
- äžè²«æ§ã®ãªãïŒç»åãµã€ãºããã©ãŒããããç»è³ªã®ã°ãã€ãã
- äžåè¡¡ïŒããŒã¿ã»ããå ã®ã¯ã©ã¹ãŸãã¯ã«ããŽãªã®äžåçãªååžã
ããŒã¿ã®ååŠçææ³
ããŒã¿ã®ååŠçã§æåã«è¡ãã¹ãããã® 1 ã€ã¯ããµã€ãºå€æŽã§ããäžéšã®ã¢ãã«ã¯å¯å€å ¥åãµã€ãºãåŠçããããã«èšèšãããŠããŸãããå€ãã®ã¢ãã«ã§ã¯äžè²«ããå ¥åãµã€ãºãå¿ èŠã§ããç»åã®ãµã€ãºãå€æŽãããšãç»åãåäžã«ãªããèšç®ã®è€éãã軜æžãããŸãã
ç»åã®ãµã€ãºå€æŽ
次ã®æ¹æ³ã䜿çšããŠãç»åã®ãµã€ãºãå€æŽã§ããŸãã
- ãã€ãªãã¢è£éïŒæãè¿ã4ã€ã®ãã¯ã»ã«å€ã®å éå¹³åãåãããšã«ãã£ãŠãã¯ã»ã«å€ãæ»ããã«ããŸãã
- ãã¢ã¬ã¹ããã€ããŒïŒå¹³ååããã«æãè¿ããã¯ã»ã«ã®å€ãå²ãåœãŠããããç»åã¯ãããã¯ç¶ã«ãªããŸãããèšç®é床ã¯éããªããŸãã
ãµã€ãºå€æŽã®äœæ¥ãç°¡åã«ããã«ã¯ã次ã®ããŒã«ã䜿çšã§ããŸãã
- OpenCV: ç»ååŠçã®ããã®åºç¯ãªæ©èœãåãããäžè¬çãªã³ã³ãã¥ãŒã¿ããžã§ã³ã©ã€ãã©ãªã
- PILïŒãããŒïŒïŒPython ç»åãã¡ã€ã«ãéããæäœããä¿åããããã®ç»åã©ã€ãã©ãªã
YOLOv8 ã«é¢ããŠãã¢ãã«åŠç¿äžã®ãimgszããã©ã¡ãŒã¿ã¯ãæè»ãªå ¥åãµã€ãºãå¯èœã«ãããç¹å®ã®ãµã€ãºãäŸãã°640ã«èšå®ãããå Žåãã¢ãã«ã¯å ã®ã¢ã¹ãã¯ãæ¯ãç¶æãããŸãŸãå ¥åç»åã®æ倧寞æ³ã640ãã¯ã»ã«ã«ãªãããã«ãªãµã€ãºããã
ã¢ãã«ãšããŒã¿ã»ããã®ç¹å®ã®ããŒãºãè©äŸ¡ããããšã§ããµã€ãºå€æŽãå¿ èŠãªååŠçæé ã§ãããã©ããããŸãã¯ã¢ãã«ãããŸããŸãªãµã€ãºã®ç»åãå¹ççã«åŠçã§ãããã©ãããå€æã§ããŸãã
ãã¯ã»ã«å€ã®æ£èŠå
å¥ã®ååŠçææ³ã¯æ£èŠåã§ããæ£èŠåã¯ãã¯ã»ã«å€ãæšæºç¯å²ã«ã¹ã±ãŒãªã³ã°ããããããã¬ãŒãã³ã°äžã®åæãé«éåãããã¢ãã«ã®ããã©ãŒãã³ã¹ãåäžããŸããããã§ã¯ãäžè¬çãªæ£èŠåææ³ãããã€ã玹ä»ããŸãã
- æå°-æ倧ã¹ã±ãŒãªã³ã°ïŒãã¯ã»ã«å€ã0ãã1ã®ç¯å²ã«ã¹ã±ãŒãªã³ã°ããã
- Zã¹ã³ã¢æ£èŠåïŒå¹³åå€ãšæšæºåå·®ã«åºã¥ããŠãã¯ã»ã«å€ãã¹ã±ãŒãªã³ã°ããŸãã
ã«é¢ã㊠YOLOv8ã®å Žåãæ£èŠåã¯ãã¢ãã«ã®ãã¬ãŒãã³ã°äžã«ååŠçãã€ãã©ã€ã³ã®äžéšãšããŠã·ãŒã ã¬ã¹ã«åŠçãããŸãã YOLOv8 ã¯ãRGB ãžã®å€æããã¯ã»ã«å€ã®ç¯å² [0, 1] ãžã®ã¹ã±ãŒãªã³ã°ãå®çŸ©æžã¿ã®å¹³åå€ãšæšæºåå·®å€ã䜿çšããæ£èŠåãªã©ãããã€ãã®ååŠçæé ãèªåçã«å®è¡ããŸãã
ããŒã¿ã»ããã®åå²
ããŒã¿ãã¯ãªãŒã³ã¢ããããããããŒã¿ã»ãããåå²ããæºåãæŽããŸããããŒã¿ããã¬ãŒãã³ã° ã»ãããæ€èšŒã»ãããããã³ãã¹ã ã»ããã«åå²ããŠãæªç¥ã®ããŒã¿ã§ã¢ãã«ãè©äŸ¡ããæ±åã®ããã©ãŒãã³ã¹ãè©äŸ¡ã§ããããã«ããããã«è¡ãããŸããäžè¬çãªåå²ã¯ããã¬ãŒãã³ã°ã« 70%ãæ€èšŒã« 20%ããã¹ãã« 10% ã§ããscikit-learn ã TensorFlow.
ããŒã¿ã»ãããåå²ãããšãã¯ã次ã®ç¹ãèæ ®ããŠãã ããã
- ããŒã¿ååžã®ç¶æïŒãã¬ãŒãã³ã°ã»ãããæ€èšŒã»ããããã¹ãã»ããã«ãããŠãã¯ã©ã¹ã®ããŒã¿ååžãç¶æãããŠããããšã確èªããã
- ããŒã¿æŒæŽ©ã®åé¿ïŒéåžžãããŒã¿å¢åŒ·ã¯ããŒã¿ã»ãããåå²ãããåŸã«è¡ããããæ€èšŒã»ããããã¹ãã»ããããã®æ å ±ãã¢ãã«åŠç¿ã«åœ±é¿ãäžããªãããã«ãããŒã¿å¢åŒ·ããã®ä»ã®ååŠçã¯ãã¬ãŒãã³ã°ã»ããã«ã®ã¿é©çšãããã¹ãã§ããã-ã¯ã©ã¹ã®ãã©ã³ã¹ããšãïŒäžåè¡¡ãªããŒã¿ã»ããã«å¯ŸããŠã¯ãå°æ°ã¯ã©ã¹ããªãŒããŒãµã³ããªã³ã°ããããå€æ°ã¯ã©ã¹ãã¢ã³ããŒãµã³ããªã³ã°ããããããã¯ããã¯ãæ€èšããã
ããŒã¿æ¡åŒµãšã¯?
æãäžè¬çã«è°è«ãããããŒã¿ã®ååŠçã¹ãããã¯ãããŒã¿æ¡åŒµã§ããããŒã¿æ¡åŒµã¯ãç»åã®ä¿®æ£ããŒãžã§ã³ãäœæããããšã§ãããŒã¿ã»ããã®ãµã€ãºã人çºçã«å¢ãããŸããããŒã¿ãæ¡åŒµããããšã§ãéåŠç¿ãæžãããã¢ãã«ã®æ±åãæ¹åã§ããŸãã
ããŒã¿æ¡åŒµã®ãã®ä»ã®å©ç¹ã¯æ¬¡ã®ãšããã§ãã
- ããããã¹ããªããŒã¿ã»ãããäœæïŒããŒã¿å¢åŒ·ã¯ãå ¥åããŒã¿ã®å€åãæªã¿ã«å¯ŸããŠã¢ãã«ãããããã¹ãã«ããŸããããã«ã¯ãç §æãæ¹åãã¹ã±ãŒã«ã®å€åãå«ãŸããŸãã
- è²»çšå¯Ÿå¹æïŒããŒã¿å¢åŒ·ã¯ãæ°ããããŒã¿ãåéãã©ãã«ä»ãããããšãªãããã¬ãŒãã³ã°ããŒã¿ã®éãå¢ããè²»çšå¯Ÿå¹æã®é«ãæ¹æ³ã§ããã
- ããŒã¿ã®æå¹æŽ»çšïŒæ°ããªããªãšãŒã·ã§ã³ãçã¿åºãããšã§ãå©çšå¯èœãªãã¹ãŠã®ããŒã¿ãã€ã³ããæ倧éã«æŽ»çšããã
ããŒã¿æ¡åŒµææ³
äžè¬çãªãªãŒã°ã¡ã³ããŒã·ã§ã³ææ³ã«ã¯ãå転ãå転ãæ¡å€§çž®å°ãè²èª¿æŽãªã©ããããŸããAlbumentationsãImgaugãããã³ TensorFlowã® ImageDataGenerator ã¯ããããã®æ¡åŒµãçæã§ããŸãã
YOLOv8 ã«é¢ããŠã¯ã.yaml ãã¡ã€ã«ã§ããããŒã¿ã»ããèšå®ãã¡ã€ã«ãä¿®æ£ããããšã§ãã«ã¹ã¿ã ã»ããŒã¿ã»ãããæ¡åŒµããããšãã§ããããã®ãã¡ã€ã«ã«ãã©ã®ããã«ããŒã¿ãè£åŒ·ããããæå®ãããã©ã¡ãŒã¿ãŒãå«ãè£åŒ·ã»ã¯ã·ã§ã³ãè¿œå ããããšãã§ããã
Ultralytics YOLOv8 ãªããžããªã¯ãå¹ åºãããŒã¿æ¡åŒµããµããŒãããŠããŸãã次ã®ãããªããŸããŸãªå€æãé©çšã§ããŸãïŒ
- ã©ã³ãã ã¯ããã
- å転:ç»åãæ°Žå¹³ãŸãã¯åçŽã«å転ã§ããŸãã
- å転:ç»åãç¹å®ã®è§åºŠã§å転ãããããšãã§ããŸãã
- æªã¿
ãŸããç¹å®ã®ãã©ã¡ãŒã¿ãŒã䜿çšããŠãããã®æ¡åŒµææ³ã®åŒ·åºŠã調æŽããããå€ãã®ããŒã¿ã®çš®é¡ãçæã§ããŸãã
ååŠçã®äºäŸç 究
以äžã䜿çšããŠã亀éç»åå ã®ããŸããŸãªçš®é¡ã®è»äž¡ãæ€åºããŠåé¡ããã¢ãã«ã®éçºãç®çãšãããããžã§ã¯ããèããŠã¿ãŸãããã YOLOv8.亀éæ å ±ã®ç»åãåéããããŠã³ãã£ã³ã°ããã¯ã¹ãšã©ãã«ã§æ³šéãä»ããŸããã
ãã®ãããžã§ã¯ãã®ååŠçã®åã¹ãããã¯æ¬¡ã®ããã«ãªããŸãã
- ç»åã®ãµã€ãºå€æŽ:以é YOLOv8 æè»ãªå ¥åãµã€ãºãåŠçãããµã€ãºå€æŽãèªåçã«å®è¡ãããããæåã§ãµã€ãºãå€æŽããå¿ èŠã¯ãããŸãããã¢ãã«ã¯ããã¬ãŒãã³ã°äžã«æå®ããã 'imgsz' ãã©ã¡ãŒã¿ãŒã«åŸã£ãŠç»åãµã€ãºã調æŽããŸãã
- ãã¯ã»ã«å€ã®æ£èŠå: YOLOv8 ååŠçäžã«ãã¯ã»ã«å€ã 0 ãã 1 ã®ç¯å²ã«èªåçã«æ£èŠåãããããå¿ é ã§ã¯ãããŸããã
- ããŒã¿ã»ããã®åå²: scikit-learn ãªã©ã®ããŒã«ã䜿çšããŠãããŒã¿ã»ããããã¬ãŒãã³ã° (70%)ãæ€èšŒ (20%)ããã¹ã (10%) ã®ã»ããã«åå²ããŸãã
- ããŒã¿æ¡åŒµ: ããŒã¿ã»ããæ§æãã¡ã€ã« (.yaml) ãå€æŽããŠãã©ã³ãã ãªããªãã³ã°ãæ°Žå¹³æ¹åã®å転ãæããã®èª¿æŽãªã©ã®ããŒã¿æ¡åŒµææ³ãå«ããŸãã
ãããã®æé ã«ãããããŒã¿ã»ãããæœåšçãªåé¡ãªãã§æºåãããæ¢çŽ¢çããŒã¿åæ (EDA) ã®æºåãæŽã£ãŠããããšã確èªãããŸãã
æ¢çŽ¢çããŒã¿è§£æææ³
ããŒã¿ã»ããã®ååŠçãšæ¡åŒµãçµãã£ããã次ã®ã¹ãããã¯æ¢çŽ¢çããŒã¿è§£æãéããŠã€ã³ãµã€ããåŸãããšã§ããEDA ã§ã¯ãçµ±èšçææ³ãšèŠèŠåããŒã«ã䜿çšããŠãããŒã¿ã®ãã¿ãŒã³ãšååžãç解ããŸããã¯ã©ã¹ã®äžåè¡¡ãå€ãå€ãªã©ã®åé¡ãç¹å®ããããŒã¿ã®ååŠçãã¢ãã« ãã¬ãŒãã³ã°ã®èª¿æŽã«ã€ããŠååãªæ å ±ã«åºã¥ãã決å®ãäžãããšãã§ããŸãã
çµ±èšçEDAææ³
çµ±èšææ³ã¯ãå€ãã®å Žåãå¹³åãäžå€®å€ãæšæºåå·®ãç¯å²ãªã©ã®åºæ¬çãªææšãèšç®ããããšããå§ãŸããŸãããããã®ææšã¯ããã¯ã»ã«åŒ·åºŠååžãªã©ãç»åããŒã¿ã»ããã®ããããã£ã®æŠèŠããã°ããæäŸããŸãããããã®åºæ¬çãªçµ±èšãç解ããããšã§ãããŒã¿ã®å šäœçãªå質ãšç¹æ§ãææ¡ããäžèŠåæ§ãæ©æã«çºèŠããããšãã§ããŸãã
ããžã¥ã¢ã«EDAã®ãã¯ããã¯
èŠèŠåã¯ãç»åããŒã¿ã»ããã® EDA ã®éµãšãªããŸããããšãã°ãã¯ã©ã¹ã®äžåè¡¡åæã¯ãEDAã®ããäžã€ã®éèŠãªåŽé¢ã§ããããã¯ãããŒã¿ã»ããã§ç¹å®ã®ã¯ã©ã¹ãéå°è©äŸ¡ãããŠãããã©ãããå€æããã®ã«åœ¹ç«ã¡ãæ£ã°ã©ãã䜿çšããŠããŸããŸãªç»åã¯ã©ã¹ãŸãã¯ã«ããŽãªã®ååžãèŠèŠåãããšãäžåè¡¡ããã°ããæããã«ããããšãã§ããŸããåæ§ã«ãå€ãå€ã¯ããã¯ã»ã«åŒ·åºŠãç¹åŸŽååžã®ç°åžžã匷調ããç®±ã²ãå³ãªã©ã®èŠèŠåããŒã«ã䜿çšããŠç¹å®ã§ããŸããå€ãå€æ€åºã¯ãç°åžžãªããŒã¿ãã€ã³ããçµæãæªããã®ãé²ããŸãã
èŠèŠåã®äžè¬çãªããŒã«ã«ã¯ã次ã®ãã®ããããŸãã
- ãã¹ãã°ã©ã ãšç®±ã²ãå³: ãã¯ã»ã«å€ã®ååžãææ¡ããå€ãå€ãç¹å®ããã®ã«åœ¹ç«ã¡ãŸãã
- æ£åžå³: ç»åã®ç¹åŸŽã泚ééã®é¢ä¿ã調ã¹ãã®ã«åœ¹ç«ã¡ãŸãã
- ããŒãããã: ãã¯ã»ã«åŒ·åºŠã®ååžããç»åå ã®æ³šéä»ãç¹åŸŽã®ç©ºéååžãèŠèŠåããã®ã«å¹æçã§ãã
äœ¿çš Ultralytics Explorer for EDA
EDA ãžã®ããé«åºŠãªã¢ãããŒãã«ã€ããŠã¯ã Ultralytics ãšã¯ã¹ãããŒã©ãŒ ããŒã«ãã³ã³ãã¥ãŒã¿ãŒ ããžã§ã³ ããŒã¿ã»ãããæ¢çŽ¢ããããã®å ç¢ãªæ©èœãæäŸããŸãããã®ããŒã«ã¯ãã»ãã³ãã£ãã¯æ€çŽ¢ãSQL ã¯ãšãªããã¯ãã«é¡äŒŒæ€çŽ¢ããµããŒãããããšã§ãããŒã¿ã®åæãšç解ã容æã«ããŸãã㧠Ultralytics ãšã¯ã¹ãããŒã©ãŒã§ã¯ãããŒã¿ã»ããã®åã蟌ã¿ãäœæããŠãé¡äŒŒã®ç»åãèŠã€ããããSQLã¯ãšãªãå®è¡ããŠè©³çŽ°ãªåæãè¡ã£ãããã»ãã³ãã£ãã¯æ€çŽ¢ãå®è¡ãããããããšãã§ããŸãã
æãå·®ã䌞ã¹ãŠã€ãªãã
ä»ã®ã³ã³ãã¥ãŒã¿ãŒ ããžã§ã³æ奜家ãšãããžã§ã¯ãã«ã€ããŠè©±ãåãããšã§ãããŸããŸãªèŠç¹ããæ°ããã¢ã€ãã¢ãåŸãããŸããããã§ã¯ãåŠç¿ããã©ãã«ã·ã¥ãŒãã£ã³ã°ããããã¯ãŒã¯ãæ§ç¯ããããã®åªããæ¹æ³ãããã€ã玹ä»ããŸãã
ã³ãã¥ããã£ãšã€ãªããããã®ãã£ã³ãã«
- GitHub IssuesïŒ YOLOv8 GitHub ãªããžããªã«ã¢ã¯ã»ã¹ããIssues ã¿ãã䜿ã£ãŠè³ªåãæãããããããã°ãå ±åããããæ©èœãææ¡ãããããŠãã ãããã³ãã¥ããã£ãšã¡ã³ãããŒã¯ãããªããçŽé¢ããã©ããªåé¡ã«ã察å¿ããŸãã
- Ultralytics DiscordãµãŒããŒïŒ Ultralytics Discord ãµãŒããŒã«åå ããŠãä»ã®ãŠãŒã¶ãŒãéçºè ãšã€ãªããããµããŒããåããç¥èãå ±æããã¢ã€ãã¢ãåºãåããŸãããã
å ¬åŒææž
- Ultralytics YOLOv8 ããã¥ã¡ã³ãæ°å€ãã®ã³ã³ãã¥ãŒã¿ããžã§ã³ã¿ã¹ã¯ãšãããžã§ã¯ãã«é¢ãã培åºçãªã¬ã€ããšè²ŽéãªæŽå¯ã«ã€ããŠã¯ã YOLOv8 ã®å ¬åŒããã¥ã¡ã³ããåç §ããŠãã ããã
ããŒã¿ã»ããã®æºåãæŽããŸããã
ããŒã¿ã®ãµã€ãºå€æŽãæ£èŠåãæ¡åŒµããŒã¿ãé©åã«è¡ããšããã€ãºãäœæžãããæ±åãæ¹åãããã¢ãã«ã®ããã©ãŒãã³ã¹ãåäžããŸãããã®ã¬ã€ãã§æŠèª¬ãããŠããååŠçã®ææ³ãšãã¹ã ãã©ã¯ãã£ã¹ã«åŸãããšã§ããœãªãã ããŒã¿ã»ãããäœæã§ããŸããååŠçãããããŒã¿ã»ããã®æºåãã§ããããèªä¿¡ãæã£ãŠãããžã§ã¯ãã®æ¬¡ã®ã¹ãããã«é²ãããšãã§ããŸãã
ããããã質å
ã³ã³ãã¥ãŒã¿ããžã§ã³ãããžã§ã¯ãã«ãããããŒã¿ååŠçã®éèŠæ§ãšã¯ïŒ
ããŒã¿ã®ååŠçã¯ãããŒã¿ãã¯ãªãŒã³ã§äžè²«æ§ããããã¢ãã«åŠç¿ã«æé©ãªåœ¢åŒã§ããããšãä¿èšŒãããããã³ã³ãã¥ãŒã¿ããžã§ã³ãããžã§ã¯ãã«ãããŠäžå¯æ¬ ã§ãããçããŒã¿ã®ãã€ãºãäžæŽåãäžåè¡¡ãªã©ã®åé¡ã«å¯ŸåŠããããšã§ããµã€ãºå€æŽãæ£èŠåãæ¡åŒµãããŒã¿ã»ããåå²ãªã©ã®ååŠçã¹ãããã¯ãèšç®è² è·ã軜æžããã¢ãã«æ§èœãåäžãããã®ã«åœ¹ç«ã¡ãŸãã詳现ã«ã€ããŠã¯ãã³ã³ãã¥ãŒã¿ããžã§ã³ãããžã§ã¯ãã®ã¹ããããã芧ãã ããã
Ultralytics YOLO ãããŒã¿è£åŒ·ã«äœ¿ãã«ã¯ïŒ
Ultralytics YOLOv8 ã§ããŒã¿ãæ¡åŒµããã«ã¯ãããŒã¿ã»ããèšå®ãã¡ã€ã«ïŒ.yamlïŒãä¿®æ£ããå¿ èŠãããããã®ãã¡ã€ã«ã§ã¯ãã©ã³ãã ã¯ããããæ°Žå¹³å転ãæãã調æŽãªã©ãããŸããŸãªãªãŒã°ã¡ã³ããŒã·ã§ã³æè¡ãæå®ã§ãããããã¯ãããã§èª¬æãããã¬ãŒãã³ã°ã»ã³ã³ãã£ã®ã¥ã¬ãŒã·ã§ã³ã䜿ã£ãŠå¹æçã«è¡ãããšãã§ãããããŒã¿å¢åŒ·ã¯ãããããã¹ããªããŒã¿ã»ãããäœæãããªãŒããŒãã£ããã£ã³ã°ãæžãããã¢ãã«ã®äžè¬åãæ¹åããã®ã«åœ¹ç«ã¡ãŸãã
ã³ã³ãã¥ãŒã¿ããžã§ã³ããŒã¿ã«æé©ãªããŒã¿æ£èŠåæè¡ãšã¯ïŒ
ããŒãã©ã€ãŒãŒã·ã§ã³ã¯ããã¯ã»ã«å€ãæšæºçãªç¯å²ã«ã¹ã±ãŒãªã³ã°ããããšã§ãåæãæ©ãããã¬ãŒãã³ã°æã®ããã©ãŒãã³ã¹ãåäžãããŸããäžè¬çãªææ³ã«ã¯ä»¥äžãå«ãŸããŸãïŒ
- æå°-æ倧ã¹ã±ãŒãªã³ã°ïŒãã¯ã»ã«å€ã0ãã1ã®ç¯å²ã«ã¹ã±ãŒãªã³ã°ããã
- Zã¹ã³ã¢æ£èŠåïŒå¹³åå€ãšæšæºåå·®ã«åºã¥ããŠãã¯ã»ã«å€ãã¹ã±ãŒãªã³ã°ããŸãã
YOLOv8 ã«ã€ããŠã¯ãRGBãžã®å€æããã¯ã»ã«å€ã®ã¹ã±ãŒãªã³ã°ãªã©ãæ£èŠåãèªåçã«åŠçãããã詳ããã¯ã¢ãã«ãã¬ãŒãã³ã°ã®ã»ã¯ã·ã§ã³ãã芧ãã ããã
ãã¬ãŒãã³ã°çšã®ã¢ãããŒã·ã§ã³ããŒã¿ã»ããã¯ã©ã®ããã«åå²ããã°ããã®ã§ããããïŒ
ããŒã¿ã»ãããåå²ããã«ã¯ã70%ããã¬ãŒãã³ã°çšã20%ãæ€èšŒçšã10%ããã¹ãçšã«åããã®ãäžè¬çã§ãããããã®åå²ã«ããã£ãŠã¯ã©ã¹ã®ããŒã¿ååžãç¶æãããã¬ãŒãã³ã°ã»ããã«å¯ŸããŠã®ã¿ãªãŒã°ã¡ã³ããŒã·ã§ã³ãå®è¡ããããšã§ããŒã¿æŒããåé¿ããããšãéèŠã§ãããããŒã¿ã»ããåå²ãå¹ççã«è¡ãã«ã¯ãscikit-learn ãTensorFlow ã®ãããªããŒã«ã䜿çšãããããŒã¿ã»ããã®æºåã«é¢ãã詳现ãªã¬ã€ããåç §ããŠãã ããã
æåã§ãªãµã€ãºããªããŠããYOLOv8 ãããŸããŸãªç»åãµã€ãºã«å¯Ÿå¿ã§ããŸããïŒ
ã¯ããUltralytics YOLOv8 ãã¢ãã«ã®ãã¬ãŒãã³ã°äžã«'imgsz'ãã©ã¡ãŒã¿ãéããŠãæ§ã ãªç»åãµã€ãºãæ±ãããšãã§ããŸãããã®ãã©ã¡ãŒã¿ã¯ãã¢ã¹ãã¯ãæ¯ãç¶æããªãããç»åã®æ倧寞æ³ãæå®ããããµã€ãºïŒäŸãã°640ãã¯ã»ã«ïŒã«åãããã«ãªãµã€ãºãããããšãä¿èšŒããŸããããæè»ãªå ¥ååŠçãšèªå調æŽã«ã€ããŠã¯ãã¢ãã«ãã¬ãŒãã³ã°ã®ã»ã¯ã·ã§ã³ãã芧ãã ããã