I-Intel ishicilele ukukhishwa kwephrojekthi ye-oidn 2.1 (Open Image Denoise), ethuthukisa iqoqo lezihlungi zezithombe ezikhipha umsindo ezilungiselelwe kusetshenziswa amasistimu okunikezwa kwe-ray tracing. I-Open Image Denoise ithuthukiswa njengengxenye yephrojekthi enkulu ye-API Rendering Toolkit ehloselwe ukuthuthukisa amathuluzi okubona esofthiwe yesayensi (SDVis (Software Defined Visualization)), okuhlanganisa nelabhulali yokulandelela i-Embree ray, isistimu yokunikezela ngezithombe ze-GLuRay, iplathifomu esabalalisiwe ye-OSRay yokulandelela ama-ray, kanye ne-OpenSWR software+2.0 ishicilelwe ngaphansi kwelayisensi ye-Apache.
Umgomo wephrojekthi ukuhlinzeka ngezici zekhwalithi ephezulu, eziphumelelayo, nezisebenziseka kalula ze-denoising ezingase zisetshenziswe ukuthuthukisa ikhwalithi yemiphumela yokulandela imisebe. Izihlungi ezihlongozwayo zivumela, ngokusekelwe emphumeleni womjikelezo wokulandela umsebe omfushane, ukuthola ileveli yekhwalithi yokugcina eqhathaniswa nomphumela wenqubo ebiza kakhulu nedla isikhathi yokunikezwa okunemininingwane.
Vula i-Image Denoise ihlunga umsindo ongahleliwe njengokulandelela imisebe ye-Monte Carlo Numerical Integration (MCRT). Ukuze kuzuzwe ukunikezwa kwekhwalithi ephezulu kuma-algorithms anjalo, ukulandelelwa kwenani elikhulu kakhulu lemisebe kuyadingeka, ngaphandle kwalokho ama-artifact abonakalayo ngendlela yomsindo ongahleliwe avela esithombeni esiwumphumela.
Ukusetshenziswa kwe-Open Image Denoise kukuvumela ukuthi wehlise inani lezibalo ezidingekayo ngama-oda amaningana wobukhulu lapho ubala iphikseli ngayinye. Ngenxa yalokho, kungenzeka ukukhiqiza isithombe esinomsindo ekuqaleni ngokushesha okukhulu, kodwa bese usiletha kukhwalithi eyamukelekayo usebenzisa ama-algorithms okunciphisa umsindo ngokushesha. Ngezisetshenziswa ezifanele, amathuluzi ahlongozwayo angasetshenziswa ngisho nasekulandeleleni imisebe ehlangene ngokususa umsindo lapho undiza.
Umtapo wolwazi ungasetshenziswa ezigabeni ezahlukahlukene zamadivayisi, ukusuka kumalaptop nama-PC kuye kumanodi ngamaqoqo. Ukuqaliswa kwenzelwe amakilasi ahlukahlukene e-Intel CPUs angu-64-bit asekela imiyalo ye-SSE4, AVX2, AVX-512 kanye ne-XMX (Xe Matrix Extensions), ama-Apple Silicon chips namasistimu ane-Intel Xe (Arc, Flex and Max series), i-NVIDIA (esekelwe ku-Volta, Turing, Ampere, Ada Lovelace ne-Hopper kanye ne-AMD2 architectures (DNANavix21) esekelwe ku-RD3 ne-AMD3 (DNANavix4.1) ) izakhiwo) ). Ukusekelwa kwe-SSEXNUMX kumenyezelwe njengemfuneko encane.
Izinguquko ezibalulekile ekukhishweni kwe-Open Image Denoise 2.1:
- Kwengezwe ukusekelwa kokukhipha umsindo kwezithombe ngesiteshi esisodwa (esibonisa ngale) noma ezimbili (idatha esobala nokujula) iziteshi ezengeziwe.
- Usekelo olungeziwe lwenhlanganisela engafanele yezinhlobo zedatha esithombeni sokufaka. Isibonelo, i-OIDN_FORMAT_FLOAT3 yemibala kanye ne-OIDN_FORMAT_HALF3 ye-albedo.
- Ukuthuthukiswa kokusebenza kwenzelwe iningi lezakhiwo ze-GPU.
- Usekelo olubuyiselwe lwenketho yokwakha ye-OIDN_STATIC_LIB kumaskripthi e-CMake, okukuvumela ukuthi wakhe amalabhulali amile (i-CPU-kuphela) kanye ne-hybrid (CPU ne-GPU).
- Indlela yokukhipha() yengezwe ezintweni ze-C++ API (DeviceRef, BufferRef, FilterRef).
- Ibhekane nenkinga yamazinga aphansi ekhwalithi kuma-Intel ahlanganisiwe we-GPU ngezindleko zokusebenza.
- Kungezwe ukusekelwa kwe-Intel oneAPI DPC++/C++ Compiler 2024.0 kanye nezinguqulo zomthombo ovulekile ezihambisanayo.
Source: opennet.ru
