Enkathini yobuhlakani bokwenziwa nokufunda okujulile, i-PyTorch iwumthombo ovulekile wolwazi wokufunda womshini ovulekile wePython onokuhlanganisa kwe-tensor kanye namanethiwekhi ajulile e-neural. Esinye sezici zayo eziningi eziwusizo yi-PyTorchVideo, okuyithuluzi eliklanyelwe ngokukhethekile imisebenzi yokuqonda ividiyo. Kulesi sihloko, sizongena emhlabeni we-PyTorchVideo, izinkinga engasisiza ukuthi sibhekane nazo, futhi sikuhambise ekusetshenzisweni kwayo.
I-PyTorchVideo: Uhlolojikelele olufushane
Ividiyo ye-PyTorch ingumtapo wolwazi othuthukiswe yi-Facebook AI, edalelwe ukusiza abacwaningi nonjiniyela ekwakheni amamodeli okuqonda amavidiyo asebenza kahle kakhulu. Ilabhulali iqukethe izingxenye ezifana nezilayishi zedathasethi yevidiyo, amamodeli aqeqeshwe kusengaphambili okuqonda ividiyo, namathuluzi wamamethrikhi nokuhlola. Nge-PyTorchVideo, kuba lula ukusebenza ngedatha yevidiyo futhi uthuthukise ukunemba kwemisebenzi yokuqonda ividiyo efana nokuhlukanisa, ukutholwa kwezinto, nokuningi.
Ukubhekana Nezinkinga Zokuqonda Ngevidiyo
Izinkinga zokuqonda ividiyo zingaba yinselele impela, ngenxa yenani lomthamo wedatha kumavidiyo, uma kuqhathaniswa nezithombe. Le nkimbinkimbi yenza ukuqeqeshwa nokucubungula amamodeli okuqonda amavidiyo kudle isikhathi esiningi futhi kushube ngokwekhompyutha. I-PyTorchVideo ifuna ukuxazulula lezi zinkinga ngokunikeza i-ecosystem ebanzi yemisebenzi yokuqonda ividiyo nokwenza ifinyeleleke kakhudlwana kubathuthukisi.
Manje ake singene ekusetshenzisweni kwe-PyTorchVideo kanye nomhlahlandlela wesinyathelo ngesinyathelo wokuthi usetshenziswa kanjani.
Isinyathelo 1: Kubalulekile ukuthi i-PyTorch ifakwe ngaphambi kokusebenzisa i-PyTorchVideo. Indlela elula yokuyithola ngokusebenzisa i-pip:
pip install torch torchvision
Isinyathelo 2: Faka i-PyTorchVideo ngokusebenzisa umyalo olandelayo:
pip install pytorchvideo
Ilayisha amasethi edatha evidiyo
Esinye sezici ezibalulekile ezinikezwe i-PyTorchVideo yikhono lokusebenza namasethi edatha evidiyo ahlukahlukene. Ake sihlole indlela yokulayisha idathasethi yesampula kusetshenziswa Imojula Yedatha Ye-Kinetics.
from pytorchvideo.data import KineticsDataModule # Configure the dataloader data_config = { "train_path": "path/to/train/dataset", "val_path": "path/to/validation/dataset", "batch_size": 8, } # Initializing the DataModule kinetics_data_module = KineticsDataModule.from_config_dict(data_config)
Lokhu kuzolayisha idathasethi ye-Kinetics, engasetshenziswa ukuqeqesha nokuqinisekisa amamodeli akho okuqonda amavidiyo.
Ukusebenza namamodeli aqeqeshwe ngaphambilini
I-PyTorchVideo ihlinzeka ngamamodeli ahlukahlukene aqeqeshwe ngaphambilini emisebenzi yokuqonda ividiyo. Lawa mamodeli angasetshenziswa njengoba enjalo kweminye imisebenzi, noma ashunwe kahle ukuze afinyelele ukusebenza okungcono kudathasethi yevidiyo yakho ethile. Nasi isibonelo sendlela yokulayisha imodeli eqeqeshwe ngaphambilini.
from pytorchvideo.models import slowfast # Load a pre-trained SlowFast model slowfast_model = slowfast.slowfast_r50()
Kafushane, i-PyTorchVideo iyilabhulali enamandla ngendlela emangalisayo eyenza imisebenzi yokuqonda ividiyo ibe lula ngokunikeza izilayishi zedathasethi, amamodeli aqeqeshwe kusengaphambili, namathuluzi awusizo wamamethrikhi nokuhlola. Ngaleli thuluzi, abathuthukisi bangakha kalula amamodeli okuqonda amavidiyo asebenza kahle kakhulu futhi anembe, abe neqhaza ekuthuthukisweni komkhakha wobuhlakani bokwenziwa nokufunda okujulile. Ngakho-ke qhubeka uhlole umhlaba we-PyTorchVideo ukuze uthathe amaphrojekthi akho okuqonda ividiyo uwayise ezingeni elilandelayo.