我们需要一个额外的 Python 依赖项

!pip install timm

import torch
# list of models: 'mealv1_resnest50', 'mealv2_resnest50', 'mealv2_resnest50_cutmix', 'mealv2_resnest50_380x380', 'mealv2_mobilenetv3_small_075', 'mealv2_mobilenetv3_small_100', 'mealv2_mobilenet_v3_large_100', 'mealv2_efficientnet_b0'
# load pretrained models, using "mealv2_resnest50_cutmix" as an example
model = torch.hub.load('szq0214/MEAL-V2','meal_v2', 'mealv2_resnest50_cutmix', pretrained=True)
model.eval()

所有预训练模型都期望输入图像以相同方式归一化，即形状为 (3 x H x W) 的三通道 RGB 图像 mini-batches，其中 H 和 W 预计至少为 224。图像必须加载到 [0, 1] 范围，然后使用 mean = [0.485, 0.456, 0.406] 和 std = [0.229, 0.224, 0.225] 进行归一化。

以下是示例执行。

# Download an example image from the pytorch website
import urllib
url, filename = ("https://github.com/pytorch/hub/raw/master/images/dog.jpg", "dog.jpg")
try: urllib.URLopener().retrieve(url, filename)
except: urllib.request.urlretrieve(url, filename)

# sample execution (requires torchvision)
from PIL import Image
from torchvision import transforms
input_image = Image.open(filename)
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_batch = input_tensor.unsqueeze(0) # create a mini-batch as expected by the model

# move the input and model to GPU for speed if available
if torch.cuda.is_available():
    input_batch = input_batch.to('cuda')
    model.to('cuda')

with torch.no_grad():
    output = model(input_batch)
# 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)

# Download ImageNet labels
!wget https://raw.githubusercontent.com/pytorch/hub/master/imagenet_classes.txt

# Read the categories
with open("imagenet_classes.txt", "r") as f:
    categories = [s.strip() for s in f.readlines()]
# Show top categories per image
top5_prob, top5_catid = torch.topk(probabilities, 5)
for i in range(top5_prob.size(0)):
    print(categories[top5_catid[i]], top5_prob[i].item())

模型描述

MEAL V2 模型来自论文 MEAL V2: Boosting Vanilla ResNet-50 to 80%+ Top-1 Accuracy on ImageNet without Tricks。

在本文中，我们介绍了一种简单而有效的方法，可以将原版 ResNet-50 在 ImageNet 上的 Top-1 准确率提升到 80%+，而且不使用任何技巧。总的来说，我们的方法基于最近提出的 MEAL，即通过判别器进行集成知识蒸馏。我们通过以下方式进一步简化它：1) 仅在最终输出上采用相似性损失和判别器；2) 使用所有教师集成的 softmax 概率平均值作为更强的监督进行蒸馏。我们方法的一个关键视角是，在蒸馏过程中不应使用 one-hot/hard 标签。我们表明，这样一个简单的框架可以实现最先进的结果，而无需涉及任何常用技巧，例如：1) 修改架构；2) 使用 ImageNet 之外的额外训练数据；3) autoaug/randaug；4) 余弦学习率调度；5) mixup/cutmix 训练；6) 标签平滑等。

模型 分辨率 参数数量 Top-1/Top-5  
MEAL-V1 w/ ResNet50 224 25.6M 78.21/94.01 GitHub
MEAL-V2 w/ ResNet50 224 25.6M 80.67/95.09  
MEAL-V2 w/ ResNet50 380 25.6M 81.72/95.81  
MEAL-V2 + CutMix w/ ResNet50 224 25.6M 80.98/95.35  
MEAL-V2 w/ MobileNet V3-Small 0.75 224 2.04M 67.60/87.23  
MEAL-V2 w/ MobileNet V3-Small 1.0 224 2.54M 69.65/88.71  
MEAL-V2 w/ MobileNet V3-Large 1.0 224 5.48M 76.92/93.32  
MEAL-V2 w/ EfficientNet-B0 224 5.29M 78.29/93.95  

参考文献

更多详情请参阅我们的论文 MEAL V2 和 MEAL。

@article{shen2020mealv2,
    title={MEAL V2: Boosting Vanilla ResNet-50 to 80%+ Top-1 Accuracy on ImageNet without Tricks},
    author={Shen, Zhiqiang and Savvides, Marios},
    journal={arXiv preprint arXiv:2009.08453},
    year={2020}
}

@inproceedings{shen2019MEAL,
	title = {MEAL: Multi-Model Ensemble via Adversarial Learning},
	author = {Shen, Zhiqiang and He, Zhankui and Xue, Xiangyang},
	booktitle = {AAAI},
	year = {2019}
}