圖像預(yù)訓(xùn)練模型的起源解說和使用示例

來源：Deephub Imba
本文約3500字，建議閱讀14分鐘
本文文章簡要介紹了研究人員在圖像識別算法和圖像數(shù)據(jù)方面的演變，并總結(jié)了現(xiàn)在的一些熱門話題。

• ImageNet
• 預(yù)訓(xùn)練模型
• 遷移學(xué)習(xí)（熱門話題）
• 使用預(yù)訓(xùn)練模型識別未知圖像
• PyTorch

ImageNet 的起源

什么是預(yù)訓(xùn)練模型？

1989 年，Yann LeCun 等人。在貝爾實(shí)驗(yàn)室結(jié)合了一個(gè)由反向傳播算法訓(xùn)練的卷積神經(jīng)網(wǎng)絡(luò)來讀取手寫數(shù)字，并成功地將其應(yīng)用于識別美國郵政服務(wù)提供的手寫郵政編碼號碼。所以此處應(yīng)該有一張LeCun老師的照片

遷移學(xué)習(xí)技術(shù)

使用預(yù)訓(xùn)練模型識別未知圖像

import ioimport torchfrom PIL import Imageimport requestsfrom torch.autograd import Variableimport torchvision.models as modelsimport torchvision.transforms as transforms
# You can get all the pre-trained models here: https://pytorch.org/vision/stable/models.htmlvgg16 = models.vgg16(pretrained=True) # This may take a few minutes.vgg16

with open("/Downloads/imagenet_classes.txt", "r") as f:  labels = [s.strip() for s in f.readlines()]

import urllib#url, filename = ("https://github.com/dataman-git/codes_for_articles/blob/master/pic/tesla.png?raw=true", "tesla.jpg")url, filename = ("https://cff2.earth.com/uploads/2022/01/06080341/Goldfish.jpg?raw=true", "goldfish.jpg")url, filename = ("https://cdn.britannica.com/92/152292-050-EAF28A45/Bald-eagle.jpg?raw=true", "eagle.jpg")
try: urllib.URLopener().retrieve(url, filename)except: urllib.request.urlretrieve(url, filename)
# sample execution (requires torchvision)from PIL import Imagefrom torchvision import transformsinput_image = Image.open(filename).convert('RGB')print(input_image.size)input_image.show()

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)input_tensor.shape # = torch.Size([3, 224, 224])input_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the modelinput_tensor.shape # = torch.Size([1, 3, 224, 224])
# move the input and model to GPU for speed if availableif torch.cuda.is_available():  input_batch = input_batch.to('cuda')  model.to('cuda')

output = vgg16(input_batch) # Returns a Tensor of shape (batch, num class labels)
# Tensor of shape 1000, with confidence scores over Imagenet's 1000 classes#print(output[0])# The output has unnormalized scores. To get probabilities, you can run a softmax on it.probabilities = torch.nn.functional.softmax(output[0], dim=0)#print(probabilities)

# Show top categories per imagetop5_prob, top5_catid = torch.topk(probabilities, 5)for i in range(top5_prob.size(0)):  print(labels[top5_catid[i]], top5_prob[i].item())

總結(jié)