InoteveraGaidhi rinotevera
Kurasikirwa Kwetatu uye Metric Kudzidza
Basics
Nhungamiro yehunyanzvi
Siamese Networks uye Triplet Kurasikirwa
Siamese network anoshandisa maviri kana anopfuura akafanana, ekugovanisa huremu matavi kuti adzidze maitiro maviri akafanana ekuisa, pane kurongedza rimwe nerimwe.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
Triplet loss trains them by pulling matching items together and pushing mismatches apart, which is the backbone of face recognition, signature verification, and one-shot learning.
Kudzika Kwakadzika
Iyo Siamese network inomhanyisa ega ega encoder ine huremu hwakagovaniswa, ichigadzira yekumisikidza vector kune yega yega. Panzvimbo pekufanotaura kirasi label, inofananidza embeddings uchishandisa chinhambwe senge Euclidean kana cosine. Izvi zvinoita kuti sisitimu ione zvikwata zvitsva zvayasina kumbodzidzira pazviri - zvakakosha kana uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge uchinge? Shanduro dzekutanga dzakashandisa kurasikirwa kwakasiyana pavaviri (zvakafanana vs. dissimilar). Kurasikirwa katatu kwakavandudza izvi nekudzidziswa pane matatu ekuisa kamwechete: anchor, yakanaka (yakafanana kirasi seancho), uye isina kunaka (yakasiyana kirasi). Chinangwa chinomanikidza chinhambwe cheanchor-positive kuti ive diki pane chinhambwe cheanchor-negative nechekumucheto, saka modhi inodzidza nzvimbo yekumisikidza apo zvinhu zvakafanana-zviziviso zvakaungana zvakasimba uye zvitupa zvakasiyana zvinogara zvakaparadzana.
Kuitwa muNyika yeChokwadi
Kuzivikanwa kwechiso pamafoni (FaceNet-maitiro): simbisa chitupa nekutarisa kana maviri akamisikidzwa kumeso ari pedyo zvakakwana.
Siginicha uye simbiso yerunyoro, ichisimbisa kana sampuli ichienderana nereferensi pafaira.
Kuonekwa kwakapetwa uye padhuze-kudzokororwa, kuwana mapikicha anooneka akafanana echigadzirwa kana mifananidzo yakatorwa.
Kudzidzira kumwe-kupfura kwezvikamu zvisingawanzo, kuziva munhu mutsva kana chinhu kubva kune imwechete yakanyoreswa muenzaniso.
Ramangwana reSiamese Networks uye Triplet Kurasikirwa
Pfungwa yepakati - dzidza nzvimbo yekumisikidza iyo chinhambwe chakaenzana - ikozvino inotyaira yakakura-yakasiyana kudzidza. Nzira dzakaita seSimCLR uye modhi dzakaita seCLIP dzinojairisa musimboti mumwe chete kumamirioni emifananidzo uye mavara maviri pasina akajeka katatu. Tarisira kudzidza kwemetric kuti irambe iri pakati pekutora, kudhirowa, kurudziro, uye vector-database yekutsvaga, nepo kurasikirwa kutsva (InfoNCE, kuwanda-kufanana) uye mabheji makuru ari kuwedzera kutsiva ruoko-tuned katatu migodhi yekubudirira uye chiyero.
Kuzvishandisa
Nei zvichikosha
- Mutengo uye bhajeti. Zvisarudzo zvezvivakwa zvinotyaira kuita uye mutengo wekushandisa kwemakore.
- Sarudzo dzakajeka. Dzidzo yehunyanzvi inobatsira zvikwata kusarudza murwi wakakodzera, kwete iwo mutsva chete.
- Kudzora kwemhando yepamusoro. Sarudzo dzeinjiniya dziri nani dzinoderedza zviitiko zvekuvimbika mukugadzira.
Chenjera
- Kugadzirisa imwe bhenji kunogona kuvanza yakafara system kushaya simba.
- Infrastructure uye mari yekugadzirisa inowanzotarisirwa pasi.
- Chengetedzo uye kucherechedzwa mapundu anogona kukura sezvo masisitimu anowedzera kuoma.
Matanho anotevera
- Tsanangura latency, mhando, uye mutengo zvinangwa usati waitwa.
- Benchmark pasi pechokwadi mutoro uye data mamiriro.
- Chishandiso chekutarisa zvikanganiso, kudonha, uye mushandisi maitiro.
- Gadzirira nzira dzekudzosera kumashure uye dzezviitiko usati wawedzera.
Ramba Uchiongorora
Tsamba yemahara
Tora iyo yemazuva ese AI pfupiso
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Take the Siamese Networks and Triplet Loss quiz
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
Mibvunzo inowanzo bvunzwa
What is Siamese Networks and Triplet Loss?
Siamese network anoshandisa maviri kana anopfuura akafanana, ekugovanisa huremu matavi kuti adzidze maitiro maviri akafanana ekuisa, pane kurongedza rimwe nerimwe. Kurasikirwa katatu kunovadzidzisa nekukwevera zvinhu zvinofananidzira pamwe chete uye kusundira zvisizvo zvakasiyana, inova musana wekuzivikanwa kwechiso, siginecha yekuongorora, uye pfuti imwe chete kudzidza.
Chii chinotsanangudza chivakwa cheSiamese network?
Mapazi ekugovanisa huremu akafanana anovimbisa kuti zvese zvinopinda zvakaiswa mepu munzvimbo imwechete yekumisikidza kuitira kuti madaro aenzaniswe.
Mukurasikirwa katatu, ndezvipi zvitatu zvekupinza?
Kurasikirwa katatu kunoshandisa anchor, yakanaka (yakafanana kirasi) uye isina kunaka (yakasiyana kirasi) kugadzira nzvimbo yekumisikidza.
Chinangwa chekurasikirwa katatu chinosimbisa chii?
Kurasikirwa kunosundidzira chakanaka pedyo kupfuura chakaipa nechero muganho, kubatanidza zvinhu zvekirasi imwechete pamwechete.
Sei 'hard-negative mining' yakakosha pakudzidzisa kurasikirwa katatu?
Kana isina kunaka yatove kure, kurasikirwa kuri zero; kusarudza nekunyengera zvisizvo pedyo zvinopa muenzaniso chimwe chinhu chine musoro kudzidza.
Ndeupi hunyanzvi hunoita kuti maSiamese network akoshese 'pakudzidza-kamwe chete'?
Nemhaka yokuti vanodzidza kufanana panzvimbo pemavara akaiswa, vanogona kuziva chiziviso chitsva kubva mumuenzaniso mumwe chete wakanyoreswa.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi