µ¼¶Á£ºÔÚCVPR 2020ÉÏ£¬£¬£¬£¬£¬£¬ÎªÁ˸üºÃµÄ½â¾öÎïÌå¼ì²âÖеıê×¼ÎÊÌ⣬£¬£¬£¬£¬£¬ÉÌÌÀEIGËã·¨ÖÐ̨ÍŶÓÖØÐÂÉè¼ÆÁ˾­µäµÄµ¥½×¶Î¼ì²âÆ÷µÄFPN^¡¾1¡¿ÒÔ¼°HEAD½á¹¹£¬£¬£¬£¬£¬£¬Í¨¹ý½á¹¹¸ü¾ßµÈ±äÐÔµÄÌØÕ÷½ð×ÓËþ£¬£¬£¬£¬£¬£¬ÒÔÌá¸ß¼ì²âÆ÷Ó¦¶Ô±ê׼ת±äµÄ³°ôÐÔ£¬£¬£¬£¬£¬£¬¿ÉÒÔʹµ¥½×¶Î¼ì²âÆ÷ÔÚcocoÉÏÌáÉý¡«4mAP£¬£¬£¬£¬£¬£¬ÍêÕû´úÂëÒÑ¿ªÔ´¡£¡£¡£¡£¡£¡£¡£

                    ͼ 1 Ч¹û¸ÅÀÀ

ÄîÍ·

ÎÄÕÂÌá³öÁËÒ»¸öÕë¶ÔÎïÌå¾ßÓдó±ê׼ת±äÊý¾Ý¼¯µÄ¼ì²âËã·¨¡£¡£¡£¡£¡£¡£¡£ÊÂÇéµÄÑо¿ÄîÍ·ÔÚÓÚ£º

1.     Ä¿½ñËùÓлùÓÚRetinaNetµÄµ¥½×¶Î¼ì²âÆ÷£¬£¬£¬£¬£¬£¬¶¼½ÓÄÉÁ˹²ÏíȨÖصķÖÀàHEADÒÔ¼°»Ø¹éHEADÔÚÌØÕ÷½ð×ÖËþÉÏ»¬¶¯´Ó¶ø¼ì²â²î±ð¾ÞϸµÄÎïÌ壬£¬£¬£¬£¬£¬ÏìÓ¦µÄ×îΪ˳ӦÕâÖֽṹµÄÌØÕ÷½ð×ÓËþ¹ØÓÚ±ê׼ת±äÓ¦µ±¾ßÓеȱäÐÔ¡£¡£¡£¡£¡£¡£¡£

2.     ¾­µäµÄµ×²ãÌØÕ÷ÌáÈ¡Ëã·¨ÈçSIFT£¬£¬£¬£¬£¬£¬Æä¸ß˹½ð×ÖËþ¾ßÓкܺõıê×¼µÈ±äÐÔ£¬£¬£¬£¬£¬£¬ÎÒÃÇ˼Ë÷ÁËÉî¶ÈÉñ¾­ÍøÂçÖÐÌØÕ÷½ð×ÖËþÓë¸ß˹½ð×ÖËþµÄÇø±ð£¬£¬£¬£¬£¬£¬´Ó¶øÌá³öÁËÌá¸ßÆäµÈ±äÐÔµÄËã·¨¡£¡£¡£¡£¡£¡£¡£

¹ØÓÚ¸ß˹½ð×ÖËþµÄµÈ±äÐÔ×Ó£¬£¬£¬£¬£¬£¬¿ÉÒÔÍŽáÏÂÃæÒ»¸öʾÀýΪ¸÷ÈËչʾÕâÖÖÐÔ×Ó£º

  ͼ 2            ͼ3

    £º¶ÔÑò×ì±ßÑؾßÓиßÏìÓ¦µÄÌØÕ÷ÌáÈ¡Ëã×Ó £º´ødilationµÄÑò×ì±ßÑØÌáÈ¡Ëã×Ó

ͼ2ͼ3ÖУ¬£¬£¬£¬£¬£¬Ò»ÕÅͼÖÐÓÐÁ½Ö»±ê×¼±ÈÀý»òÐíΪ2µÄÑò£¬£¬£¬£¬£¬£¬ÓÐÒ»¸ö¿ÉÒÔ¼ì²âÑò×ìµÄÌØÕ÷ÌáÈ¡Æ÷£¬£¬£¬£¬£¬£¬¿ÉÊÇÆä½öÄܶԱê×¼½ÏСµÄÑòµÄÑò×ì¾ßÓнϸߵÄÏìÓ¦£¬£¬£¬£¬£¬£¬ÄÇô¹ØÓÚ±ê×¼½Ï´óµÄÑòÎÒÃÇÓ¦¸ÃÈçÄÇÀïÖÃÄØ£¿£¿£¿£¿£¿£¿£¿

1.     Èçͼ2£¬£¬£¬£¬£¬£¬¿ÉÒÔÓÃÒ»¸ö¸ß˹¾í»ýºËÈ¥³ý¸ßƵºó½µ²ÉÑùÒ»´Î£¬£¬£¬£¬£¬£¬Ê¹µÃ´óÑòµÄÑò×ìÌØÕ÷¸úСÑòÔÚͳһ±ê×¼£¬£¬£¬£¬£¬£¬´Ó¶øÔÚ½µ²ÉÑùµÄͼÉϾÙÐдóÑòÑò×ìµÄ¼ì²â¡£¡£¡£¡£¡£¡£¡£

2.     Èçͼ3£¬£¬£¬£¬£¬£¬¿ÉÒÔÓÃÒ»¸ö¸ß˹¾í»ýºËÈ¥³ý¸ßƵºó£¬£¬£¬£¬£¬£¬½«ÌØÕ÷ÌáÈ¡Ëã×Ó±äΪ±äΪÆÓª¾í»ý£¬£¬£¬£¬£¬£¬Òà¿ÉÒÔ¾ÙÐдóÑòµÄÑò×ì¼ì²â¡£¡£¡£¡£¡£¡£¡£

ÕâÁ½ÖÖÒªÁì²»µ«ÇкÏÖ±¾õ£¬£¬£¬£¬£¬£¬ÆäÔÚÊý×ÖÐźŴ¦Öóͷ£ÀíÂÛÖÐÓнÏΪÍêÉƵÄÀíÂÛÂÛÖ¤£¬£¬£¬£¬£¬£¬Ôڴ˲»ÔÙ׸Êö¡£¡£¡£¡£¡£¡£¡£

×·Äîµ¥½×¶Î¼ì²âÆ÷£¨ÈçRetinaNet£©£¬£¬£¬£¬£¬£¬

 Í¼ 4 retinanet½á¹¹

Æä¹²ÏíȨÖصÄHEAD(Subnet)Éè¼Æ×ÅʵÒѾ­½«ÌØÕ÷½ð×ÖËþÖеÄÌØÕ÷¿´³ÉÁ×Æß˹½ð×ÖËþ£¬£¬£¬£¬£¬£¬¼´¼ÙÉèÌØÕ÷½ð×ÓËþÒ²ÓÐÉÏÊö±ê×¼µÈ±äÐÔ×Ó¡£¡£¡£¡£¡£¡£¡£¿ÉÊÇÏÖʵʹÓÃÖУ¬£¬£¬£¬£¬£¬ÎÒÃǾ­³£ÊӲ쵽²î±ð±ê×¼µÄÎïÌåÏ໥¾ºÕù,ʹµÃ¼ì²âÆ÷ÔÚ²î±ð±ê×¼µÄÐÔÄܴ˸߱˵͵ÄÕ÷Ï󣬣¬£¬£¬£¬£¬ÕâЩÇéÐβ»Ì«Çкϱê×¼µÈ±äÌØÕ÷£¬£¬£¬£¬£¬£¬ÕâÊÇΪʲôÄØ£¿£¿£¿£¿£¿£¿£¿

ÆÊÎö

Ïà±È¸ß˹½ð×ÖËþ£¬£¬£¬£¬£¬£¬´ÓbackboneµÄ²î±ðstage³éÈ¡ÌØÕ÷ͼ×é³ÉµÄµÄÌØÕ÷½ð×ÖËþ×ÅʵʧȥÁËÑÏ¿áµÄµÈ±äÐÔ¡£¡£¡£¡£¡£¡£¡£

1      Ïà±ÈÓÚͼÏñ´¦Öóͷ£ÖÐlow levelµÄÌØÕ÷£¬£¬£¬£¬£¬£¬Éî¶ÈÉñ¾­ÍøÂçÌáÈ¡µÄÌØÕ÷ÍùÍù¾­ÓÉ×ã¹»¶à»ù´¡ÌØÕ÷µÄ×éºÏ£¬£¬£¬£¬£¬£¬ÆäÌØÕ÷ÍùÍùÓиüΪÖØ´óµÄ¸ß¼¶ÓïÒ壬£¬£¬£¬£¬£¬²¢ÇÒÓÉÓÚÎÒÃÇÓÃÏàͬµÄbackbone¾ÙÐÐÌØÕ÷µÄÌáÈ¡²î±ð±ê×¼µÄÎïÌåµÄÌØÕ÷»áÓÐÏ໥¾ºÕùµÄÕ÷Ï󡣡£¡£¡£¡£¡£¡£

2      Éî¶ÈÉñ¾­ÍøÂçÖУ¬£¬£¬£¬£¬£¬ÓÉÓÚ×ã¹»¶àµÄ¾í»ýÓë·ÇÏßÐÔ¼¤»îµ¥Î»£¬£¬£¬£¬£¬£¬Æä²î±ðstage¸ÐÊÜÒ°µÄת±äÔÚÌØÕ÷ͼµÄ²î±ðλÖþßÓз×ÆçÖµÄÕ÷Ï󣬣¬£¬£¬£¬£¬Ê¹µÃ½ð×ÓËþµÄÁ½²ã¼äºÜÄÑÓиß˹½ð×ÖËþÒ»Ñù£¬£¬£¬£¬£¬£¬²î±ð¾ÞϸÎïÌåµÄÌØÕ÷ÔÚdownsampleÒ»´Î£¬£¬£¬£¬£¬£¬»òÕßÔö´óÌØÕ÷ÌáÈ¡Ëã×ÓµÄdilationÕâÑùµÄ´¦Öóͷ£ºó¾ßÓÐÒ»ÖÂÐÔ¡£¡£¡£¡£¡£¡£¡£

¹ØÓÚ²î±ð±ê×¼ÎïÌåµÄ¾ºÕù£º

                                                      ͼ 5

ÎïÌå¼ì²âÖУ¬£¬£¬£¬£¬£¬ÎÒÃÇÐèÒªÌØÕ÷ͼͬʱ¾ßÓнϸߵÄÓïÒåÐÅÏ¢ÍêÒòËØÀ࣬£¬£¬£¬£¬£¬ÓÖÐèÒª½ÏºÃµÄ±£´æԭͼÐÅÏ¢ÓÃÓÚ¶¨Î»£¬£¬£¬£¬£¬£¬¼òµ¥±ê×¼µÄÎïÌ壬£¬£¬£¬£¬£¬ÔÚbackboneÖеIJî±ðstage¿ÉÒÔÌáÈ¡³ö²î±ð¼¶±ðµÄÓïÒ壬£¬£¬£¬£¬£¬¿ÉÊÇÔÚ´¦Öóͷ£ÁíÒ»±ê×¼µÄÎïÌåʱ£¬£¬£¬£¬£¬£¬Ôò»á·ºÆð²»Æ¥ÅäµÄÕ÷Ï󣬣¬£¬£¬£¬£¬ÈçСÎïÌåÓõ½µÄÌØÕ÷ÌáÈ¡Ëã×Ó¿ÉÄÜÖ»ÊÇ´óÎïÌåÌáÈ¡³õ¼¶ÌØÕ÷Ëã×Ó£¬£¬£¬£¬£¬£¬´Ó¶øºÜÄÑÌáÈ¡¸ß¼¶ÓïÒå¾ÙÐзÖÀ࣬£¬£¬£¬£¬£¬ÕâÒ²Æô·¢ÁËØʺóµÄFPNµÄTop-down½á¹¹£¬£¬£¬£¬£¬£¬ÒÔ¼°ºóÐøµÄһϵÁÐfeature fusingµÄ½á¹¹¡£¡£¡£¡£¡£¡£¡£

ÍøÂçÉè¼Æ

ÔõÑùË¢ÐÂÌØÕ÷½ð×ÓËþʹÆä¾ßÓиüºÃµÄµÈ±äÐÔ×ÓÄØ£¿£¿£¿£¿£¿£¿£¿

ÎÒÃÇ·¢Ã÷£¬£¬£¬£¬£¬£¬ÔÚÌØÕ÷½ð×ÓËþµÄijһ²ãΪ»ù×¼£¬£¬£¬£¬£¬£¬ÆäÓëÉÏÏÂÁ½²ã¼ä¾ßÓиüºÃµÄÏà¹ØÐÔ£¬£¬£¬£¬£¬£¬Ö±¾õÉÏ¿ÉÒÔÍƲâÆäµ×²¿Ò»²ãÍùÍù±£´æÓиü¶àµÄ¶¨Î»Ïà¹ØµÄÐÅÏ¢£¬£¬£¬£¬£¬£¬¸ü¸ß²ãÍùÍù¾ßÓиüºÃµÄ·ÖÀàÏà¹ØµÄÐÅÏ¢£¬£¬£¬£¬£¬£¬¶øÒÔÍùµÄfeature fusingµÄÒªÁ춼ÒÅ©ÁËÕâÖÖ¿Õ¼ä±ê×¼£¨ÌØÕ÷½ð×ÓËþ²ã¼ä£©µÄÏà¹ØÐÔ¡£¡£¡£¡£¡£¡£¡£

                   Í¼ 6  FPN²î±ð²ã¼äµÄÏà¹ØÐÔ

       ÎÞÂÛÊÇÐÎʽÉϵÄÆÊÎöÕÕ¾ÉʵÑé·¢Ã÷µÄÕâÖÖÏà¹ØÐÔÌåÏÖÎÒÃDZê×¼¿Õ¼äÉÏȷʵÓÐÖµµÃÌáÈ¡µÄÌØÕ÷£¬£¬£¬£¬£¬£¬ÎÒÃÇÏ£Íûͨ¹ýÔÚ±ê×¼¿Õ¼äÉϾÙÐоí»ýÀ´ÌáÈ¡ÕâÖÖÌØÕ÷¡£¡£¡£¡£¡£¡£¡£

       ˼Á¿µ½FPNµÄ²î±ð²ãµÄÇø·ÖÂʲî±ð£¬£¬£¬£¬£¬£¬ÎÒÃÇÓÉ´ËÌá³ö½ð×ÖËþ¾í»ý£¨Pconv£©µÄ½á¹¹¡£¡£¡£¡£¡£¡£¡£

       ÌØÕ÷½ð×ÖËþÏàÁÚÈý²ã£¨P3,P4,P5£©¾ÙÀý£¬£¬£¬£¬£¬£¬ÔÚP3ÉÏʹÓÃstrideΪ2µÄconv¾ÙÐоí»ý£¬£¬£¬£¬£¬£¬ÔÚP4ÉÏʹÓÃͨË×conv¾í»ý£¬£¬£¬£¬£¬£¬P5ÉÏͨË×conv¾í»ýÈ»ºóupsample£¬£¬£¬£¬£¬£¬»ñµÃÏàͬ¾ÞϸµÄÌØÕ÷ͼȻºóÏà¼Ó£¬£¬£¬£¬£¬£¬²»ÄÑÍƵ¼£¬£¬£¬£¬£¬£¬ÆäµÈЧÓÚÔÚÌØÕ÷ͼÉÏÿһµãÔÚHWά¶È¾í»ýºóÔÙÔÚ±ê×¼¿Õ¼ä£¨P3£¬£¬£¬£¬£¬£¬P4£¬£¬£¬£¬£¬£¬P5£©¾ÙÐÐÒ»´Î¾í»ý£¬£¬£¬£¬£¬£¬Òò´ËÆäÏ൱ÓÚÒ»¸ö3-DµÄ¾í»ýºË¡£¡£¡£¡£¡£¡£¡£

 Í¼ 7 ±ê×¼¿Õ¼ä¾í»ý£¨Pconv£© ʾÒâͼ

¼ÈÈ»ÓÐÁ˱ê×¼¿Õ¼äµÄ¾í»ý£¬£¬£¬£¬£¬£¬ÎÒÃÇ×ÔÈ»µÄÏëµ½ÁË¿Õ¼ä±ê×¼¹²ÏíËùÓвÎÊýµÄÅú¹éÒ»»¯£¨iBN£©£¬£¬£¬£¬£¬£¬ÎÒÃÇÿһ´ÎPconvºó£¬£¬£¬£¬£¬£¬ÎÒÃǽ«ËùÓвãÅþÁ¬ÔÚÒ»Æ𣬣¬£¬£¬£¬£¬ÅäºÏ¸üÐÂͳ¼ÆÁ¿£¬£¬£¬£¬£¬£¬ÕâÒ»ÉèÅÌËãNAS-FPN^¡¾2¡¿ÖÐʹÓõĸ÷±ê×¼Óø÷×ÔµÄBN²ÎÊý¾ßÓиüÓŵÄÐÔÄÜÇÒ¸üÒ×¹¤³Ì»¯£¨ÈçÍÆÀíʱȥ³ýBN£©¡£¡£¡£¡£¡£¡£¡£

ͼ 8   iBN ʾÒâͼ

       ÁíÒ»·½Ã棬£¬£¬£¬£¬£¬¸ß˹½ð×ÓËþÖÐÌØÕ÷ÏàÁڲ㽵²ÉÑùʱÓÐÆäÏàÆ¥ÅäµÄ¸ß˹ģºýºË£¬£¬£¬£¬£¬£¬Ì«´óµÄÄ£ºýºË»áÈ¥³ý¹ý¶àÓÐÓÃÐÅÏ¢£¬£¬£¬£¬£¬£¬Ì«Ð¡µÄÄ£ºýºËÒýÈë¸ßƵÔëÉù¡£¡£¡£¡£¡£¡£¡£¿ÉÊÇÌØÕ÷½ð×ÖËþÖÐÏàÁÚ²ãÖ®¼ä¾àÀë²î±ðµÄ¾í»ýºËÓë´ó×ڵķÇÏßÐÔµ¥Î»£¬£¬£¬£¬£¬£¬ÕâʹµÃÌØÕ÷ͼÉϲî±ðλÖÃÏàÁÚ²ãÖ®¼äûÓÐÒ»¸öÀο¿µÄÀíÂÛ¸ß˹ģºýºË¡£¡£¡£¡£¡£¡£¡£Òò´ËÎÒÃÇÓÉ´Ë»ùÓÚÒÔÉÏÌá³öµÄPconvÓÖ½«ÆäË¢ÐÂΪ±ê×¼µÈ»¯µÄ±ê×¼¿Õ¼ä¾í»ý£¨SEPC£©Ä£¿£¿£¿£¿£¿£¿£¿é£ºµ±±ê×¼¿Õ¼ä¾í»ýºËÔÚ±ê×¼¿Õ¼äÉÏ»¬¶¯Ê±£¬£¬£¬£¬£¬£¬×îµ×²ãÉÏÓÃͨË×µÄ3X3¾í»ý£¬£¬£¬£¬£¬£¬µ±ÏàͬµÄ¾í»ýºË»¬Ïò¸ß²ãÌØÕ÷ͼʱÓÿɱäÐξí»ý£¨Dconv£©^¡¾3¡¿ÊµÏÖÏñÔÚ¸ß˹½ð×ÖËþÖÐÒ»ÑùµÄ¸ß²ãÌØÕ÷µãÓë×îµ×²ãµÄ¶ÔÆë¡£¡£¡£¡£¡£¡£¡£Í¬Ê±Õâ¸öÉè¼ÆÖпɱäÐξí»ýÖ»Ôڸ߲ãÌØÕ÷ͼÉϽÓÄÉ£¬£¬£¬£¬£¬£¬ÆäÒýÈëµÄÌØÁíÍâDconvµÄÅÌËãÁ¿¿ªÏúÒÔ¼°ÍÆÀíËÙÂÊËðʧ×ÅʵÊǺÜСµÄ¡£¡£¡£¡£¡£¡£¡£

ͼ 9 SEPC ʾÒâͼ

ʵÑéЧ¹û

       ÎÒÃÇ×öÁËÏ꾡µÄÏûÈÚʵÑ飬£¬£¬£¬£¬£¬ÒÔÈ·±£ÎÒÃÇÿ¸öÄ£¿£¿£¿£¿£¿£¿£¿éµÄÓÐÓÃÐÔ.ÎÒÃÇÑ¡È¡ÁËFSAF^¡¾4¡¿£¬£¬£¬£¬£¬£¬RetinaNet£¬£¬£¬£¬£¬£¬FreeAnchor^¡¾5¡¿Èý¸öÓдú±íÐÔµÄmodelÑéÖ¤¹þ¹þ(haha)ÌåÓýÓÐÓÃÐÔ¡£¡£¡£¡£¡£¡£¡£

       ¿ÉÒÔ¿´µ½£¬£¬£¬£¬£¬£¬ÎÒÃÇÌá³öµÄÄ£¿£¿£¿£¿£¿£¿£¿éÔÚÈý¸ömodelÉϾùÓÐÎȹ̵ÄÌáÉý£¬£¬£¬£¬£¬£¬²¢ÇÒÐÔÄÜÌáÉýÏà±ÈflopsÓëforwardËÙÂʵÄÔöÌí¼«¾ßÐԼ۱ȣ¬£¬£¬£¬£¬£¬ÖµµÃ×¢ÖصÄÊÇ£¬£¬£¬£¬£¬£¬ËäÈ»ÎÒÃÇʹÓÃÁËDconv£¬£¬£¬£¬£¬£¬¿ÉÊÇÐÔÄܵÄÌáÉý²»µ«½öÊÇDconv´øÀ´µÄ£¬£¬£¬£¬£¬£¬Ïà±Èhead½á¹¹ËùÓÐÌ滻ΪDconv£¬£¬£¬£¬£¬£¬¹þ¹þ(haha)ÌåÓýSEPCÔÚËÙÂÊÓëÐÔÄÜÉ϶¼ÌåÏÖ³öÁ˾ø¶ÔµÄÓÅÊÆ¡£¡£¡£¡£¡£¡£¡£

ͬʱÎÒÃÇÑ¡È¡ÁËFreeAnchorΪ»ù´¡ÓëÄ¿½ñSOTA¼ì²âÆ÷¾ÙÐÐÁ˽ÏÁ¿¡£¡£¡£¡£¡£¡£¡£

       ÎÒÃÇSEPCÔÚmAP45×óÓÒµÄbaselineÉÏÒÀÈ»Óп¿½ü3mAPµÄÌáÉý£¬£¬£¬£¬£¬£¬ÎÒÃÇÔÚÉõÖÁÔÚµ¥±ê×¼²âÊÔµÄÌõ¼þÏ»ñµÃÁËÒ»¸ömAPΪ50.1µÄµ¥½×¶Î¼ì²âÆ÷¡£¡£¡£¡£¡£¡£¡£

´«ËÍÃÅ

´úÂëÏÖÔÚÒ²ÒѾ­¿ªÔ´£¬£¬£¬£¬£¬£¬½Ó´ýÁÐλͬÑâʹÓúͽ»Á÷¡£¡£¡£¡£¡£¡£¡£

ÂÛÎĵص㣺 https://arxiv.org/pdf/2005.03101.pdf

´úÂëµØµã£º https://github.com/jshilong/SEPC

References

[1] Tsung-Yi Lin, Piotr Dollar, Ross Girshick, Kaiming He, Bharath Hariharan, and Serge Belongie. Feature pyramid networks for object detection. In CVPR, 2017.

[2]Golnaz Ghiasi, Tsung-Yi Lin, and Quoc V Le. Nas-fpn: Learning scalable feature pyramid architecture for object detection. In CVPR, 2019.

[3] Jifeng Dai, Haozhi Qi, Yuwen Xiong, Yi Li, Guodong Zhang, Han Hu, and Yichen Wei. Deformable convolutional networks. In ICCV, 2017.

[4] Chenchen Zhu, Yihui He, and Marios Savvides. Feature selective anchor-free module for single-shot object detection. In CVPR, 2019.

[5] Xiaosong Zhang, Fang Wan, Chang Liu, Rongrong Ji, and Qixiang Ye. FreeAnchor: Learning to match anchors for visual object detection. In NIPS, 2019.