Up tókànItọsọna atẹle
Perceptual Hashing for Near-Duplicate Images
AI wiwo
Visual AI Itọsọna
Isonu Iro ati LPIPS
Perceptual losses compare image representations from a neural network instead of relying only on pixel differences.
Gbọ si itọsọna yiiAI-ṣe afihan · ~2 min
Lori iwe yi2 min ka
Akopọ
LPIPS calibrates such feature distances against human similarity judgments, but no score guarantees that every image pair will be judged alike by every viewer.
Jin Dive
Awọn adanu aṣa bii L2 (aṣiṣe onigun mẹrin tumọ si) ṣe afiwe awọn aworan pixel-nipasẹ-piksẹli, nitorinaa iyipada piksẹli kan tabi awoara ti o yatọ die-die dabi aṣiṣe nla paapaa botilẹjẹpe eniyan ko ni akiyesi. Pipadanu oye dipo ṣiṣe awọn aworan mejeeji nipasẹ nẹtiwọọki iṣaaju (nigbagbogbo VGG) ati ṣe afiwe awọn iṣẹ ṣiṣe lati awọn ipele agbedemeji. Nitoripe awọn ẹya wọnyẹn ṣe koodu awọn egbegbe, awọn awoara, ati awọn ẹya ohun dipo awọn iye piksẹli deede, pipadanu naa ṣe deede dara julọ pẹlu idajọ eniyan, iwuri didasilẹ, awọn abajade otitọ ni itumọ. LPIPS (Ibajọra Aworan Patch Iro ti Kọ), ti a ṣe nipasẹ Zhang et al. ni 2018, formalizes yi: o jade jin awọn ẹya ara ẹrọ, normalizes wọn, ati ki o kan eko òṣuwọn calibrated lodi si egbegberun eda eniyan idajo, producing kan nikan ijinna Dimegilio ibi ti kekere tumo si siwaju sii perceptually bakanna.
Real-World imuse
Awọn nẹtiwọọki ipinnu-giga ikẹkọ (fun apẹẹrẹ, SRGAN) nitorina awọn fọto ti o ga soke dabi didasilẹ ati ifojuri kuku ju blurry.
Iṣirowọn funmorawon aworan ati awọn kodẹki nipa titọka bi iwoye ṣe sunmọ aworan ti a yipada si atilẹba.
Gbigbe ara itọsọna, nibiti akoonu ti baamu nipasẹ awọn ẹya VGG jin kuku ju awọn piksẹli deede.
Benchmarking GAN ati awọn olupilẹṣẹ aworan kaakiri nipasẹ jijabọ ijinna LPIPS laarin awọn aworan ti ipilẹṣẹ ati gidi.
Ojo iwaju ti Ipadanu Iro ati LPIPS
Awọn metiriki oye ti n yipada lati awọn ẹhin CNN si awọn ẹya lati abojuto ti ara ẹni ati awọn awoṣe oluyipada iran bi DINO ati CLIP, eyiti o mu awọn atunmọ ọrọ mu. Reti isọpọ titọ pẹlu ikẹkọ awoṣe-itankalẹ ati igbelewọn ọrọ-si-aworan, pẹlu awọn ikun oye ti aifwy fun aitasera akoko fidio. Awọn oniwadi tun n ṣe iwadii awọn aaye afọju LPIPS: o le jẹ aṣiwere ni ilodisi ati ailagbara ni ibamu pẹlu didara ni iṣotitọ ti o ga pupọ, ti nfa awọn metiriki ibaramu eniyan tuntun bii DISTS ati awọn isunmọ akojọpọ.
Fifi o sinu iwa
Kini idi ti o ṣe pataki
- Iyara ati iwọn. Visual AI le ṣe adaṣe adaṣe, wiwa, ati awọn iṣẹ ṣiṣe taagi ni iwọn.
- Kọ awọn yiyan. Awọn ẹgbẹ ẹda le ṣe apẹrẹ awọn imọran yiyara pẹlu awọn atunyẹwo afọwọṣe diẹ.
- Ẹgbẹ ati ṣiṣan iṣẹ. Awọn iṣẹ ṣiṣe le lo aworan ati awọn ifihan agbara fidio ti o nira tẹlẹ lati ṣiṣẹ.
Ṣọra fun
- Awọn ẹtọ aworan ati igbanilaaye le di awọn eewu labẹ ofin ti o ba jẹ afihan.
- Iṣe awoṣe le yatọ kọja ina, awọn ẹda eniyan, ati awọn agbegbe.
- Awọn idaniloju eke le ma ṣe akiyesi ayafi ti a ba ṣe abojuto awọn ala igbẹkẹle.
Next awọn igbesẹ
- Ṣetumo awọn ibeere gbigba fun pipe, iranti, ati awọn idiyele aṣiṣe.
- Ṣe idanwo pẹlu data ti o baamu awọn ipo iṣelọpọ gidi.
- Ṣafikun atunyẹwo eniyan fun igbẹkẹle kekere tabi awọn asọtẹlẹ ipa-giga.
- Tọpinpin awoṣe ki o ṣe tunṣe lẹhin kamẹra tabi awọn ayipada datasetto.
Tesiwaju Ṣiṣawari
Ìwé ìròyìn ọfẹ
Gba ìkànìyàn AI ojoojúmọ́
Ìtàn AI mẹta tó jẹ́rìí ní gbogbo owurọ́ ọjọ́ iṣẹ́, kọ́wé ní Gẹ̀ẹ́sì tó rọrùn. Ọfẹ́ fún gbogbo àkókò, kò sí ìpolówó.
Ìmeèlì kan ní gbogbo ọjọ́ iṣẹ́. Yọkúrò ní ìkànìyàn ní ìkànìyàn kan. A kò ta tàbí pín àdírẹ́sì rẹ.
Ṣe idanwo fun ara rẹ
Take the Perceptual Loss and LPIPS quiz
Ìdáhùn lẹ́sẹ̀kẹsẹ̀ lórí gbogbo ìdáhùn, àti iwe-ẹ̀rí tó lè pin pẹ̀lú ID tó lè jẹ́rìí lẹ́yìn tí o bá kọ́ ẹ̀kọ́.
Ṣe àtìlẹ́yìn ẹ̀kọ́ AI ọfẹ́. AI Understanding jẹ́ àjọ àìní èrè 501(c)(3) — kò sí ìpolówó, kò sí àkúnya owó, ní gbogbo àkókò. Ṣe ẹbun kan
Awọn ibeere ti a beere nigbagbogbo
What is Perceptual Loss and LPIPS?
Perceptual losses compare image representations from a neural network instead of relying only on pixel differences. LPIPS calibrates such feature distances against human similarity judgments, but no score guarantees that every image pair will be judged alike by every viewer.
Kini idi ti ipadanu L2 ti o da lori piksẹli nigbagbogbo ṣe aṣiṣe ibajọra aworan ni akawe si ipadanu oye?
L2 ṣe afiwe awọn piksẹli taara, nitorinaa awọn iyipada aye kekere tabi awọn iyatọ sojurigindin forukọsilẹ bi awọn aṣiṣe nla paapaa nigbati aworan ba dara si eniyan.
Kini LPIPS ni akọkọ ṣe afiwe laarin awọn aworan meji?
LPIPS yọkuro awọn iṣẹ ṣiṣe ẹya ti o jinlẹ lati ẹhin ti o wa titi ati ṣe iwọn ijinna wọn, eyiti o ṣe deede dara julọ pẹlu iwoye eniyan ju awọn piksẹli lọ.
Bawo ni a ṣe pinnu awọn iwuwo ikanni kọọkan ni LPIPS?
Awọn òṣuwọn naa ni a kọ ẹkọ lati baramu dataset nla kan (BAPPS) ti awọn ipinnu ibajọra-fipa-meji-ami-iyan eniyan.
What does a lower LPIPS value indicate for a tested image pair?
LPIPS is calibrated to human similarity judgments, but its distance remains a model-based measure with limitations.
Nẹtiwọọki ẹhin wo ni o wọpọ julọ pẹlu pipadanu oye (ẹya-ara) Ayebaye?
Awọn maapu ẹya agbedemeji VGG di boṣewa fun pipadanu oye ninu awọn iṣẹ ṣiṣe bii ipinnu-giga ati gbigbe ara.
Tesiwaju kikọ
Jẹmọ awọn itọsọna
Awọn itọsọna diẹ sii ti a yan fun koko yii