InoteveraGaidhi rinotevera
Online uye Offline Feature Kushumira Skew
Tekinoroji
Nhungamiro yehunyanzvi
Online uye Zvakaoma Negative Mining
Kuchera kwakaoma kwakashata kunotora mienzaniso inodzidzisa, yakaoma-kusiyanisa yekudzidzira pane kutambisa simba pane zviri nyore iyo modhi yatove yakarurama.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Pa peji ino2 min verenga
Pfupiso
It is the trick that makes metric learning and object detection converge fast and accurately.
Kudzika Kwakadzika
Paunenge uchidzidzira nekatatu kana kurasika kwakasiyana, mazhinji asina kujairika sampled atove kure neancho, saka anoburitsa zero kurasikirwa uye pasina gradient, matanga ekudzidzira. Negative migodhi inogadzirisa izvi nekusarudza yakaoma negatives: mienzaniso isiri iyo padyo neanchor. Mumigodhi isina Indaneti, unopota uchitarisa dhatabheti kuti uwane izvi, izvo zvinononoka uye zvinopera. Migodhi yepamhepo inovaunganidza pane nhunzi mukati meimwe mini-batch: mushure mekuenda mberi, unotarisa madaro ese ari maviri mubatch uye tora vakanyanya kutyora. FaceNet yakaunza semi-yakaoma migodhi, ichisarudza zvisina kunaka kure kupfuura iyo yakanaka asi ichiri mukati memuganho, ichidzivirira kusagadzikana uko izvo zvakaomesesa zvakashata zvinogona kukonzera kutanga mukudzidziswa.
Kuitwa muNyika yeChokwadi
Kudzidziswa kucherechedzwa kwechiso: FaceNet inoshandisa semi-yakaoma online mugodhi kudzidza embeddings inoparadzanisa vanotaridzika-vakafanana vanhu.
Kuonekwa kwechinhu: SSD uye madhigirii akafanana anoshandisa migodhi yakaoma isina kunaka kuenzanisa mafashama emabhokisi akareruka ekumashure achipokana nemabhokisi echinhu chisingawanzo.
Dense ndima yekudzoreredza: kutsvaga uye RAG masisitimu anochera magwaro asina kunaka anotaridzika akakodzera asi asiri, kurodza iyo inodzoreredza.
Masisitimu ekurudziro: mamodheru anochera zvinhu izvo mushandisi haana kudzvanya asi zvakafanana nezvinhu zvakadzvanywa, zvichidzidzisa misiyano yakanaka mukuravira.
Ramangwana reInternet uye Zvakaoma Negative Mining
Iyo musimboti, dzidzisa pane izvo zvakaoma, ikozvino inotyaira kusiyanisa-kuzvitarisisa kudzidza, uko hombe mu-batch isina kunaka madziva (uye ndangariro mabhangi seMoCo) inopa yakaoma kuenzanisa pasina mavara. Vatsvaguri vari kunatsiridza kuti kuomarara kunofanirwa kuve kwakaoma sei, sezvo zvakaomesesa-zvisina kunaka kazhinji zvinozove zvisina kunyorwa zvisizvo kana kupotsa-duplicate zvakanaka izvo zvinokanganisa kudzidziswa. Tarisira kucherwa kwakangwara, kusava nechokwadi-kuziva migodhi uye marara akaomarara anogadzirwa nemodhi pachayo, pamwe nekubatanidzwa kwakasimba nemasisitimu ekutora ayo anochera zvinonetsa kubva kumibvunzo chaiyo yemushandisi.
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 Online and Hard Negative Mining 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 Online and Hard Negative Mining?
Kuchera kwakaoma kwakashata kunotora mienzaniso inodzidzisa, yakaoma-kusiyanisa yekudzidzira pane kutambisa simba pane zviri nyore iyo modhi yatove yakarurama. Ndihwo hunyengeri hunoita kuti metric yekudzidza uye kuona kwechinhu kusanganise nekukurumidza uye nemazvo.
Nei mazhinji asina kurongeka akasarudzirwa zvakaipa achipa diki chiratidzo chekudzidzisa mukurasikirwa katatu?
Easy negatives inogara zvakanaka kupfuura muganho, inogutsa kurasikirwa nechekare, uye inopa zvakanyanya pasina gradient, saka kudzidzisa masiteki.
Chii chinosiyanisa kuchera padandemutande kubva pakuchera pasina Indaneti?
Migodhi yepamhepo inokokorodza nhambwe mbiri mukati mebatch yazvino nhanho imwe neimwe, ukuwo migodhi isina mhepo nguva nenguva ichiongorora dhata rese uye inogona kupera.
Chii chinonzi 'semi-zvakaoma' chisina kunaka sekutsanangurwa kwazvinoitwa muFaceNet?
Semi-zvakaoma zvakashata zviri kure kubva kune anchor pane yakanaka asi inowira mukati memucheto, ichipa anobatsira, akagadzikana gradients pasina kusagadzikana kweabsolute yakaoma kesi.
Mune batch-hard mining, ndeipi isina kunaka inosarudzwa pane imwe neimwe anchor?
Batch-hard mining inotora nyaya dzakaomesesa: kune imwe neimwe anchor inotora kure kure yakanaka uye iri pedyo (yakanyanya kuvhiringidza) isina kunaka mubatch.
Nei hombe batch saizi ichida kuvandudza mined metric-yekudzidza mhinduro?
Batch hombe inobereka yakakura pairwise chinhambwe matrix, saka kune akawanda evamiriri evamiriri ekuchera akaomesesa kubva padanho rega rega.
Ramba uchidzidza
Mazano anoenderana
Mamwe madhairekitori akasarudzirwa nyaya iyi