Posted 2021-04-19 鏈哄櫒瀛︿範AI绠楁硶宸ョ▼

鍚慉I杞瀷鐨勭▼搴忓憳閮藉叧娉ㄤ簡杩欎釜鍙?/span>馃憞馃憞馃憞

1.CNN鐨勭壒鐐逛互鍙婁紭鍔?/h5>

鏀瑰彉鍏ㄨ繛鎺ヤ负灞€閮ㄨ繛鎺ワ紝杩欐槸鐢变簬鍥剧墖鐨勭壒娈婃€ч€犳垚鐨勶紙鍥惧儚鐨勪竴閮ㄥ垎鐨勭粺璁＄壒鎬т笌鍏朵粬閮ㄥ垎鏄竴鏍风殑锛夛紝閫氳繃灞€閮ㄨ繛鎺ュ拰鍙傛暟鍏变韩澶ц寖鍥寸殑鍑忓皯鍙傛暟鍊笺€傚彲浠ラ€氳繃浣跨敤澶氫釜filter鏉ユ彁鍙栧浘鐗囩殑涓嶅悓鐗瑰緛锛堝鍗风Н鏍革級銆?/p>

CNN浣跨敤鑼冨洿鏄叿鏈夊眬閮ㄧ┖闂寸浉鍏虫€х殑鏁版嵁锛屾瘮濡傚浘鍍忥紝鑷劧璇█锛岃闊?/p>

1.灞€閮ㄨ繛鎺ワ細鍙互鎻愬彇灞€閮ㄧ壒寰併€?br>2.鏉冨€煎叡浜細鍑忓皯鍙傛暟鏁伴噺锛屽洜姝ら檷浣庤缁冮毦搴︼紙绌洪棿銆佹椂闂存秷鑰楅兘灏戜簡锛夈€?br>3.鍙互瀹屽叏鍏变韩锛屼篃鍙互灞€閮ㄥ叡浜紙姣斿瀵逛汉鑴革紝鐪肩潧榧诲瓙鍢寸敱浜庝綅缃拰鏍峰紡鐩稿鍥哄畾锛屽彲浠ョ敤鍜岃劯閮ㄤ笉涓€鏍风殑鍗风Н鏍革級
4.闄嶇淮锛氶€氳繃姹犲寲鎴栧嵎绉痵tride瀹炵幇銆?br>5.澶氬眰娆＄粨鏋勶細灏嗕綆灞傛鐨勫眬閮ㄧ壒寰佺粍鍚堟垚涓鸿緝楂樺眰娆＄殑鐗瑰緛銆備笉鍚屽眰绾х殑鐗瑰緛鍙互瀵瑰簲涓嶅悓浠诲姟銆?br>

2.deconv鐨勪綔鐢?/h5>

1.unsupervised learning锛氶噸鏋勫浘鍍?br>2.CNN鍙鍖栵細灏哻onv涓緱鍒扮殑feature map杩樺師鍒板儚绱犵┖闂达紝鏉ヨ瀵熺壒瀹氱殑feature map瀵瑰摢浜沺attern鐨勫浘鐗囨晱鎰?br>3.Upsampling锛氫笂閲囨牱銆?br>

3.dropout浣滅敤浠ュ強瀹炵幇鏈哄埗 (鍙傝€?https://blog.csdn.net/nini_coded/article/details/79302800)

1.dropout鏄寚鍦ㄦ繁搴﹀涔犵綉缁滅殑璁粌杩囩▼涓紝瀵逛簬绁炵粡缃戠粶鍗曞厓锛屾寜鐓т竴瀹氱殑姒傜巼灏嗗叾鏆傛椂浠庣綉缁滀腑涓㈠純銆傛敞鎰忔槸鏆傛椂锛?br>  瀵逛簬闅忔満姊害涓嬮檷鏉ヨ锛岀敱浜庢槸闅忔満涓㈠純锛屾晠鑰屾瘡涓€涓猰ini-batch閮藉湪璁粌涓嶅悓鐨勭綉缁溿€?br>2.dropout鏄竴绉岰NN璁粌杩囩▼涓槻姝㈣繃鎷熷悎鎻愰珮鏁堟灉鐨勬柟娉?br>3.dropout甯︽潵鐨勭己鐐规槸鍙兘鍑忔參鏀舵暃閫熷害锛氱敱浜庢瘡娆¤凯浠ｅ彧鏈変竴閮ㄥ垎鍙傛暟鏇存柊锛屽彲鑳藉鑷存搴︿笅闄嶅彉鎱?br>4.娴嬭瘯鏃讹紝闇€瑕佹瘡涓潈鍊间箻浠

4.娣卞害瀛︿範涓湁浠€涔堝姞蹇敹鏁?闄嶄綆璁粌闅惧害鐨勬柟娉曪細

1.鐡堕缁撴瀯
2.娈嬪樊
3.瀛︿範鐜囥€佹闀裤€佸姩閲?br>4.浼樺寲鏂规硶
5.棰勮缁?br>

5.浠€涔堥€犳垚杩囨嫙鍚堬紝濡備綍闃叉杩囨嫙鍚?/h5>

1.data agumentation
2.early stop
3.鍙傛暟瑙勫垯鍖?br>4.鐢ㄦ洿绠€鍗曟ā鍨?br>5.dropout
6.鍔犲櫔澹?br>7.棰勮缁冪綉缁渇reeze鏌愬嚑灞?br>

6.LSTM闃叉姊害寮ユ暎鍜岀垎鐐?/h5>

LSTM鐢ㄥ姞鍜岀殑鏂瑰紡鍙栦唬浜嗕箻绉紝浣垮緱寰堥毦鍑虹幇姊害寮ユ暎銆備絾鏄浉搴旂殑鏇村ぇ鐨勫嚑鐜囦細鍑虹幇姊害鐖嗙偢锛屼絾鏄彲浠ラ€氳繃缁欐搴﹀姞闂ㄩ檺瑙ｅ喅杩欎竴闂

7.涓轰粈涔堝緢澶氬仛浜鸿劯鐨凱aper浼氭渶鍚庡姞鍏ヤ竴涓狶ocal Connected Conv?

鍦ㄤ竴浜涚爺绌舵垚鏋滀腑锛屼綔鑰呴€氳繃瀹為獙琛ㄦ槑锛氫汉鑴稿湪涓嶅悓鐨勫尯鍩熷瓨鍦ㄤ笉鍚岀殑鐗瑰緛锛堢溂鐫涳紡榧诲瓙锛忓槾鐨勫垎甯冧綅缃浉瀵瑰浐瀹氾級锛屽綋涓嶅瓨鍦ㄥ叏灞€鐨勫眬閮ㄧ壒寰佸垎甯冩椂锛孡ocal-Conv鏇撮€傚悎鐗瑰緛鐨勬彁鍙栥€?/p>

8.绁炵粡缃戠粶鏉冨€煎垵濮嬪寲鏂瑰紡浠ュ強涓嶅悓鏂瑰紡鐨勫尯鍒?

鏉冨€煎垵濮嬪寲鐨勬柟娉曚富瑕佹湁锛氬父閲忓垵濮嬪寲锛坈onstant锛夈€侀珮鏂垎甯冨垵濮嬪寲锛坓aussian锛夈€乸ositive_unitball鍒濆鍖栥€佸潎鍖€鍒嗗竷鍒濆鍖栵紙uniform锛夈€亁avier鍒濆鍖栥€乵sra鍒濆鍖栥€佸弻绾挎€у垵濮嬪寲锛坆ilinear锛?/p>

9.Convolution銆?pooling銆?Normalization鏄嵎绉缁忕綉缁滀腑鍗佸垎閲嶈鐨勪笁涓楠わ紝鍒嗗埆绠€杩癈onvolution銆?pooling鍜孨ormalization鍦ㄥ嵎绉缁忕綉缁滀腑鐨勪綔鐢ㄣ€?/h5>

10.dilated conv(绌烘礊鍗风Н)浼樼己鐐逛互鍙婂簲鐢ㄥ満鏅?/h5>

鍩轰簬FCN鐨勮涔夊垎鍓查棶棰樹腑锛岄渶淇濇寔杈撳叆鍥惧儚涓庤緭鍑虹壒寰佸浘鐨剆ize鐩稿悓銆?br>鑻ヤ娇鐢ㄦ睜鍖栧眰锛屽垯闄嶄綆浜嗙壒寰佸浘size,闇€鍦ㄩ珮灞傞樁娈典娇鐢ㄤ笂閲囨牱锛岀敱浜庢睜鍖栦細鎹熷け淇℃伅锛屾墍浠ユ鏂规硶浼氬奖鍝嶅鑷寸簿搴﹂檷浣庯紱
鑻ヤ娇鐢ㄨ緝灏忕殑鍗风Н鏍稿昂瀵革紝铏藉彲浠ュ疄鐜拌緭鍏ヨ緭鍑虹壒寰佸浘鐨剆ize鐩稿悓锛屼絾杈撳嚭鐗瑰緛鍥剧殑鍚勪釜鑺傜偣鎰熷彈閲庡皬锛?br>鑻ヤ娇鐢ㄨ緝澶х殑鍗风Н鏍稿昂瀵革紝鐢变簬闇€澧炲姞鐗瑰緛鍥鹃€氶亾鏁帮紝姝ゆ柟娉曚細瀵艰嚧璁＄畻閲忚緝澶э紱
鎵€浠ワ紝寮曞叆绌烘礊鍗风Н(dilatedconvolution),鍦ㄥ嵎绉悗鐨勭壒寰佸浘涓婅繘琛?濉厖鎵╁ぇ鐗瑰緛鍥緎ize锛岃繖鏍锋棦鍥犱负鏈夊嵎绉牳澧炲ぇ鎰熷彈閲庯紝涔熷洜涓?濉厖淇濇寔璁＄畻鐐逛笉鍙樸€?br>

11.鍒ゅ埆妯″瀷鍜岀敓鎴愭ā鍨嬭В閲?/h5>

鐩戠潱瀛︿範鏂规硶鍙堝垎鐢熸垚鏂规硶锛圙enerative approach锛夊拰鍒ゅ埆鏂规硶锛圖iscriminative approach锛夛紝鎵€瀛﹀埌鐨勬ā鍨嬪垎鍒О涓虹敓鎴愭ā鍨嬶紙Generative Model锛夊拰鍒ゅ埆妯″瀷锛圖iscriminative Model锛夈€?/p>

浠庢鐜囧垎甯冪殑瑙掑害鑰冭檻锛屽浜庝竴鍫嗘牱鏈暟鎹紝姣忎釜鍧囨湁鐗瑰緛Xi瀵瑰簲鍒嗙被鏍囪yi銆?br>
鐢熸垚妯″瀷锛氬涔犲緱鍒拌仈鍚堟鐜囧垎甯働(x,y)锛屽嵆鐗瑰緛x鍜屾爣璁皔鍏卞悓鍑虹幇鐨勬鐜囷紝鐒跺悗姹傛潯浠舵鐜囧垎甯冦€傝兘澶熷涔犲埌鏁版嵁鐢熸垚鐨勬満鍒躲€?br>
鍒ゅ埆妯″瀷锛氬涔犲緱鍒版潯浠舵鐜囧垎甯働(y|x)锛屽嵆鍦ㄧ壒寰亁鍑虹幇鐨勬儏鍐典笅鏍囪y鍑虹幇鐨勬鐜囥€?br>
鏁版嵁瑕佹眰锛氱敓鎴愭ā鍨嬮渶瑕佺殑鏁版嵁閲忔瘮杈冨ぇ锛岃兘澶熻緝濂藉湴浼拌姒傜巼瀵嗗害锛涜€屽垽鍒ā鍨嬪鏁版嵁鏍锋湰閲忕殑瑕佹眰娌℃湁閭ｄ箞澶氥€?br>

鐢辩敓鎴愭ā鍨嬪彲浠ュ緱鍒板垽鍒ā鍨嬶紝浣嗙敱鍒ゅ埆妯″瀷寰椾笉鍒扮敓鎴愭ā鍨嬨€?/p>

12.濡備綍鍒ゆ柇鏄惁鏀舵暃

13.姝ｅ垯鍖栨柟娉曚互鍙婄壒鐐?/h5>

姝ｅ垯鍖栨柟娉曞寘鎷細L1 regularization 銆?L2 regularization 銆?鏁版嵁闆嗘墿澧?銆?dropout 绛?/p>

14.甯哥敤鐨勬縺娲诲嚱鏁?(鍙傝€?https://blog.csdn.net/Yshihui/article/details/80540070)

15.1x1鍗风Н鐨勪綔鐢?/h5>

1. 瀹炵幇璺ㄩ€氶亾鐨勪俊鎭氦浜掑拰鏁村悎銆?x1鍗风Н鏍稿彧鏈変竴涓弬鏁帮紝褰撳畠浣滅敤鍦ㄥ閫氶亾鐨刦eature map涓婃椂锛岀浉褰撲簬涓嶅悓閫氶亾涓婄殑涓€涓嚎鎬х粍鍚堬紝
   瀹為檯涓婂氨鏄姞璧锋潵鍐嶄箻浠ヤ竴涓郴鏁帮紝浣嗘槸杩欐牱杈撳嚭鐨刦eature map灏辨槸澶氫釜閫氶亾鐨勬暣鍚堜俊鎭簡锛岃兘澶熶娇缃戠粶鎻愬彇鐨勭壒寰佹洿鍔犱赴瀵屻€?br>2. feature map閫氶亾鏁颁笂鐨勯檷缁淬€傞檷缁磋繖涓綔鐢ㄥ湪GoogLeNet鍜孯esNet鑳藉寰堝ソ鐨勪綋鐜般€備妇涓緥瀛愶細鍋囪杈撳叆鐨勭壒寰佺淮搴︿负100x100x128锛?br>   鍗风Н鏍稿ぇ灏忎负5x5锛坰tride=1锛宲adding=2锛夛紝閫氶亾鏁颁负256锛屽垯缁忚繃鍗风Н鍚庤緭鍑虹殑鐗瑰緛缁村害涓?00x100x256锛屽嵎绉弬鏁伴噺涓?br>   128x5x5x256=819200銆傛鏃跺湪5x5鍗风Н鍓嶄娇鐢ㄤ竴涓?4閫氶亾鐨?x1鍗风Н锛屾渶缁堢殑杈撳嚭鐗瑰緛缁村害渚濈劧鏄?00x100x256锛屼絾鏄鏃剁殑鍗风Н鍙傛暟
   閲忎负128x1x1x64 + 64x5x5x256=417792锛屽ぇ绾﹀噺灏戜竴鍗婄殑鍙傛暟閲忋€?br>3. 澧炲姞闈炵嚎鎬ф槧灏勬鏁般€?x1鍗风Н鍚庨€氬父鍔犱竴涓潪绾挎€ф縺娲诲嚱鏁帮紝浣跨綉缁滄彁鍙栨洿鍔犲叿鏈夊垽鍒俊鎭殑鐗瑰緛锛屽悓鏃剁綉缁滀篃鑳藉仛鐨勮秺鏉ヨ秺娣便€?br>

16.鏃犵洃鐫ｅ涔犳柟娉曟湁鍝簺

寮哄寲瀛︿範銆並-means 鑱氱被銆佽嚜缂栫爜銆佸彈闄愭尝灏斿吂鏇兼満

17.澧炲ぇ鎰熷彈閲庣殑鏂规硶锛?/h5>

绌烘礊鍗风Н銆佹睜鍖栨搷浣溿€佽緝澶у嵎绉牳灏哄鐨勫嵎绉搷浣?br>

18.鐩爣妫€娴嬮鍩熺殑甯歌绠楁硶锛?/h5>

1.涓ら樁娈垫娴嬪櫒锛歊-CNN銆丗ast R-CNN銆丗aster R-CNN
2.鍗曢樁娈垫娴嬪櫒锛歒OLO銆乊OLO9000銆丼SD銆丏SSD銆丷etinaNet

19.鍥炲綊闂鐨勮瘎浠锋寚鏍?/h5>

1.骞冲潎缁濆鍊艰宸?MAE)
2.鍧囨柟宸?MSE)

20.鍗风Н灞傚拰鍏ㄨ繛鎺ュ眰鐨勫尯鍒?/h5>

1.鍗风Н灞傛槸灞€閮ㄨ繛鎺ワ紝鎵€浠ユ彁鍙栫殑鏄眬閮ㄤ俊鎭紱鍏ㄨ繛鎺ュ眰鏄叏灞€杩炴帴锛屾墍浠ユ彁鍙栫殑鏄叏灞€淇℃伅锛?br>2.褰撳嵎绉眰鐨勫眬閮ㄨ繛鎺ユ槸鍏ㄥ眬杩炴帴鏃讹紝鍏ㄨ繛鎺ュ眰鏄嵎绉眰鐨勭壒渚嬶紱

21.鍙嶅嵎绉殑妫嬬洏鏁堝簲鍙婅В鍐虫柟妗?/h5>

鍥惧儚鐢熸垚缃戠粶鐨勪笂閲囨牱閮ㄥ垎閫氬父鐢ㄥ弽鍗风Н缃戠粶锛屼笉鍚堢悊鐨勫嵎绉牳澶у皬鍜屾闀夸細浣垮弽鍗风Н鎿嶄綔浜х敓妫嬬洏鏁堝簲

瑙ｅ喅鏂规:

22.鍒嗙被鐨勯璁粌妯″瀷濡備綍搴旂敤鍒拌涔夊垎鍓蹭笂

1.鍙傝€冭鏂? Fully Convolutional Networks for Semantic Segmentation

23.SSD鍜孻OLO鐨勫尯鍒?/h5>

24.浜ゅ弶鐔靛拰softmax锛岃繕鏈夊畠鐨凚P

瀹炶返閮ㄥ垎

1.python涓璻ange鍜寈range鏈変粈涔堜笉鍚?/p>

涓よ€呯殑鍖哄埆鏄痻range杩斿洖鐨勬槸涓€涓彲杩唬鐨勫璞★紱range杩斿洖鐨勫垯鏄竴涓垪琛紝鍚屾椂鏁堢巼鏇撮珮锛屾洿蹇€?br>

2.python涓甫绫诲拰main鍑芥暟鐨勭▼搴忔墽琛岄『搴?/p>

1)瀵逛簬  if __name__ == '__main__': 鐨勮В閲婄浉鍏冲崥瀹㈠凡缁忕粰鍑轰簡璇存槑锛屾剰鎬濆氨鏄綋姝ゆ枃浠跺綋鍋氭ā鍧楄璋冪敤鏃讹紝涓嶄細浠庤繖閲屾墽琛岋紝
  鍥犱负姝ゆ椂name灞炴€у氨鎴愪簡妯″潡鐨勫悕瀛楋紝鑰屼笉鏄痬ain銆傚綋姝ゆ枃浠跺綋鍋氬崟鐙墽琛岀殑绋嬪簭杩愯鏃讹紝灏变細浠巑ain寮€濮嬫墽琛屻€?br>  
2)瀵逛簬甯︽湁绫荤殑绋嬪簭锛屼細鍏堟墽琛岀被鍙婄被鍐呭嚱鏁帮紝鎴栬€呭叾浠栫被澶栧嚱鏁般€傝繖閲屽彲浠ユ€荤粨涓猴紝瀵逛簬娌℃湁缂╄繘鐨勭▼搴忔锛屾寜鐓ч『搴忔墽琛屻€傜劧鍚庯紝鎵?br>  鍒癿ain鍑芥暟銆傜劧鍚庢墠鎸夌収main鍐呭嚱鏁扮殑鎵ц椤哄簭鎵ц銆傚鏋渕ain鍐呭绫昏繘琛屼簡瀹炰緥鍖栵紝閭ｄ箞鎵ц鍒版澶勬椂锛屽彧浼氬绫诲唴鎴愬憳杩涜鍒濆
  鍖栵紝鐒跺悗鍐嶈繑鍥炲埌main 鍑芥暟涓€傛墽琛屽叾浠栧疄渚嬪寲涔嬪悗瀵硅薄鐨勬垚鍛樺嚱鏁拌皟鐢ㄣ€?br>

3.绁炵粡缃戠粶鐨勫弬鏁伴噺璁＄畻

4.璁＄畻绌烘礊鍗风Н鐨勬劅鍙楅噹

5.mAP鐨勮绠?/p>

6.Python tuple鍜宭ist鐨勫尯鍒?/p>

7.Python鐨勫绾跨▼鍜屽杩涚▼锛孭ython浼绾跨▼锛屼粈涔堟椂鍊欏簲璇ョ敤瀹?/p>

8.tensorflow while_loop鍜宲ython for寰幆鐨勫尯鍒紝浠€涔堟儏鍐典笅for鏇翠紭锛?/p>

while loop鐨勫惊鐜鏁颁笉纭畾鐨勬儏鍐典笅鏁堢巼浣庯紝鍥犱负瑕佷笉鏂噸鏂板缓鍥?br>

鍙傝€冩枃鐚?/h4>

[1] https://blog.csdn.net/u014722627/article/details/77938703

[2] https://www.cnblogs.com/houjun/p/8535471.html

---

闃呰杩囨湰鏂囩殑浜鸿繕鐪嬩簡浠ヤ笅鏂囩珷锛?/strong>

涓嶆柇鏇存柊璧勬簮

娣卞害瀛︿範銆佹満鍣ㄥ涔犮€佹暟鎹垎鏋愩€乸ython

---

鏈哄ぇ鏁版嵁鎶€鏈笌鏈哄櫒瀛︿範宸ョ▼

以上是关于的主要内容，如果未能解决你的问题，请参考以下文章

谷歌浏览器调试jsp 引入代码片段，如何调试代码片段中的js

片段和活动之间的核心区别是啥？哪些代码可以写成片段？

VSCode自定义代码片段——.vue文件的模板

VSCode自定义代码片段6——CSS选择器

VSCode自定义代码片段——声明函数

VSCode自定义代码片段8——声明函数