Ixazululiwe: pytorch torchaudio torchvision cu113

Ikhaya » Python » I-Pytorch » Ixazululiwe: pytorch torchaudio torchvision cu113

I-Torchaudio kanye ne-torchvision imitapo yolwazi emibili enamandla ku-PyTorch ecosystem edlala indima ebalulekile ekucutshungulweni komsindo nemisebenzi yokubona ngekhompyutha, ngokulandelana. Kulesi sihloko, sizocubungula ngokujulile ukusebenza kwale mitapo yolwazi futhi sihlole ukuthi ingasetshenziswa kanjani ukuxazulula izinkinga eziyinkimbinkimbi emkhakheni wokucubungula idatha elalelwayo nebonwayo, sigxile enguqulweni ethi cu113. Sizophinde sixoxe ngezinyathelo zokusebenzisa la mamitapo ezincwadi ePython futhi sinikeze imininingwane ngezici zabo ezihlukile kanye namacala okusebenzisa.

Torchaudio kanye nezicelo zayo

I-torchaudio ingumtapo wolwazi we I-PyTorch ehlinzeka ngamathuluzi ahlukahlukene okucubungula umsindo, okuhlanganisa ukulayishwa kwedatha, ukuguqulwa komsindo, nokukhipha izici. Ivumela abathuthukisi ukuthi basebenzise amandla e-PyTorch ukuze baphathe idatha yomsindo futhi basebenzise ukusheshisa kwe-GPU ukuze kucutshungulwe kahle. Ezinye izinhlelo zokusebenza ezijwayelekile zifaka ukunakwa kwenkulumo, ukuhlukaniswa komsindo, nokwenza umsindo.

Ukusebenza nge-torchaudio kunembile futhi kuqondile. Okokuqala, sidinga ukufaka umtapo wolwazi uma ungekho ohlelweni lwethu. Uma ucabanga ukuthi ufake i-PyTorch, ukufakwa kwe-torchaudio kungenziwa kusetshenziswa umyalo olandelayo:

!pip install torchaudio==0.9.0 -f https://download.pytorch.org/whl/cu113/torch_stable.html

Ukuze ulayishe ifayela elilalelwayo futhi uthole i-waveform yalo nesilinganiso sesampula, singasebenzisa umsebenzi othi `torchaudio.load()`:

import torchaudio

filename = 'path/to/your/audio/file.wav'
waveform, sample_rate = torchaudio.load(filename)

I-Torchvision kanye nezicelo zayo

I-Torchvision omunye umtapo wolwazi we I-PyTorch ephethe imisebenzi yokubona ikhompuyutha ngokunikeza amasethi edatha ezithombe namavidiyo ahlukahlukene, kanye namamodeli aqeqeshwe kusengaphambili futhi aguqule ukuze kucutshungulwe izithombe. Kwenza kube lula ukudala ukuhlukaniswa kwezithombe okuyinkimbinkimbi, ukutholwa, kanye namapayipi okuhlukanisa.

Ukufaka i-torchvision, singasebenzisa umyalo olandelayo:

!pip install torchvision==0.10.0+cu113 -f https://download.pytorch.org/whl/cu113/torch_stable.html

I-Torchvision inikeza amamodeli aqeqeshwe kusengaphambili angasetshenziselwa imisebenzi ehlukene, njengokuhlukanisa izithombe. Ikhodi elandelayo ibonisa indlela yokusebenzisa imodeli eqeqeshwe kusengaphambili ukuze uhlukanise isithombe:

import torchvision.models as models
from torchvision import transforms
from PIL import Image

# Load pre-trained model
model = models.resnet18(pretrained=True)
model.eval()

# Process input image
input_image = Image.open('path/to/your/image.jpg')
preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])
input_tensor = preprocess(input_image)
batch = input_tensor.unsqueeze(0)

# Predict
output = model(batch)

Kulesi sibonelo, sisebenzise esiqeqeshwe ngaphambilini I-ResNet-18 imodeli yokuhlukaniswa kwezithombe.

Isifinyezo

Ekuphetheni, ithoshi yomsindo futhi i-torchvision (inguqulo ye-cu113) amalabhulali anamandla anweba amandla e-PyTorch, okwenza kube lula ukusebenza ngedatha yomsindo neyokubukwayo. Bavumela abathuthukisi ukuthi basebenzise izici zokufunda ezijulile kanye nokusheshisa kwe-GPU okuhlinzekwa yi-PyTorch ukuze kuxazululwe imisebenzi eyinkimbinkimbi emikhakheni yokucubungula umsindo nokubona ngekhompyutha. Sihlole ukufakwa nokusetshenziswa kwale mitapo yolwazi futhi sathinta izinhlelo zokusebenza ezivamile, ezinjengokulayisha idatha yomsindo nokuhlukaniswa kwezithombe kusetshenziswa amamodeli aqeqeshwe kusengaphambili.

Ngokuqonda nokusebenzisa le mitapo yolwazi, abathuthukisi bangathuthukisa kakhulu amakhono abo ekusebenzeni ngedatha elalelwayo nebonwayo, bavule iminyango yezixazululo ezintsha nezinhlelo zokusebenza ezisezingeni eliphezulu ekufundeni komshini kanye nobuhlakani bokwenziwa.