InoteveraGaidhi rinotevera
Monocular Depth Estimation
AI Yekutarisa
Gwara reVisual AI
DepthAnything ndiyo nheyo yemuenzaniso inofungidzira kuti kure kure kupi kwepixel yega yega kubva kune imwechete yakajairika foto, isina yakakosha hardware.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Yakagadzira yakasimba, yakajairika-chinangwa kudzika inonzwa yakachipa uye inowanikwa kune chero chinhu kubva kumafoni kusvika kumarobhoti.
DepthAnything (2024, yakaburitswa nevatsvagiri vanosanganisira avo vari TikTok/ByteDance uye HKU) inobata monocular kudzika fungidziro: kufanotaura mepu yakadzika kubva kune imwe RGB mufananidzo. Kubudirira kwayo kwaive chiyero: pachinzvimbo chekuvimba chete nedata rakanyorwa rakanyorwa rakaganhurirwa riripo, timu yakavaka injini yakanyorwa otomatiki mifananidzo inosvika 62 miriyoni isina kunyorwa ichishandisa modhi yemudzidzisi, ndokuzodzidzisa mudzidzi pane hombe iyi. Izvi zvinopa yakasimba zero-pfuti generalization mukati memukati, kunze, uye zvisingawanzo zviitiko. Izvo zvekutanga zvakabuda zvakadzika kudzika (izvo mapixels ari pedyo kana kure, kwete chaiwo metres). DepthAnything V2 (yepakati-2024) yakarodza zvinhu zvakanaka nekudzidzisa mudzidzisi nezve data rekugadzira rine chokwadi chepasi-chakanaka, ndokuzoisa kumifananidzo chaiyo, kugadzirisa micheto isina kujeka uye zvikanganiso zvinoonekera-chinhu.
Kugadzira mamepu akadzika kutyaira echokwadi kumashure blur (bokeh) mune imwe-lens smartphone portrait mafoto.
Kupa 3D chipingamupinyi maonero kune akaderera-mutengo drones uye marobhoti asina LiDAR kana stereo kamera.
Kugadzira kudzika mamiriro ekugadzirisa mamepu eControlNet saka majenareta emifananidzo anochengetedza chiitiko geometry.
Kushandura mafoto nemafirimu e2D kuita 3D kana parallax mhedzisiro yeVR uye stereoscopic kuratidza.
Tarisira kubatanidzwa kwakasimba mumagirazi eAR, makamera e-smartphone, uye marobhoti uko LiDAR yakatsaurirwa inodhura zvakanyanya kana yakakura. Metric akasiyana anoburitsa mamita echokwadi, pamwe nemavhidhiyo modhi ane kudzika kwakadzikama kwenguva (hapana kupenya pakati pemafuremu), ari kufambira mberi nekukurumidza. Sezvo aya mamodheru achidzikira kumhanya pa-mudziyo munguva chaiyo, imwe-kamera 3D maonero anozove akasarudzika kugona, kudyisa spatial komputa, yakazvimirira kufamba, uye generative 3D chiitiko kuvakazve.
Tsamba yemahara
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
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
DepthAnything ndiyo nheyo yemuenzaniso inofungidzira kuti kure kure kupi kwepixel yega yega kubva kune imwechete yakajairika foto, isina yakakosha hardware. Yakagadzira yakasimba, yakajairika-chinangwa kudzika inonzwa yakachipa uye inowanikwa kune chero chinhu kubva kumafoni kusvika kumarobhoti.
Monocular zvinoreva kudzika kunotorwa kubva kune imwe (mono) kamera mufananidzo, isina stereo pair kana inoshanda sensor.
Chikwata chakavaka injini yedata iyo pseudo-yakanyorwa makumi emamiriyoni emifananidzo isina kunyorwa, ichiwedzera zvinoshamisa chiyero chekudzidziswa.
DepthAnything inoshandisa DINOv2 ViT encoder yakapetwa neDPT-style dense yekufungidzira musoro.
Iyo base modhi inofanotaura kudzika kwakadzika kurongeka kwete yakakwana metric kureba.
V2 yakadzidzisa mudzidzisi pamifananidzo yekugadzira ine hudzamu chaihwo, yobva yanyungudutswa kumifananidzo chaiyo yemiganhu inopinza.
Ramba uchidzidza
Mamwe madhairekitori akasarudzirwa nyaya iyi
InoteveraGaidhi rinotevera
Monocular Depth Estimation
AI Yekutarisa