（表征学习论文阅读）A Simple Framework for Contrastive Learning of Visual Representations

Chen T, Kornblith S, Norouzi M, et al. A simple framework for contrastive learning of visual representations[C]//International conference on machine learning. PMLR, 2020: 1597-1607.

1. 前言

本文作者为了了解对比学习是如何学习到有效的表征，对本文所提出的三大组件进行了全面的研究：

1. 各种数据增强手段的组合在表征学习中起到了重要作用；
2. 在表征和对比损失之间引入非线性变换能够有效提高表征质量；
3. 对比学习相较于监督学习需要更大的batch size和更多的训练步数。

在没有人类标注或者监督的情况下学习数据的有效表征是一个长期存在的难题，目前的主要工作可以分为两类：

1. 基于生成模型的方法
   例如VQ-VAE,MAE,BERT
2. 基于判别模型的方法
   例如MoCo，CLIP

2. 方法

本文提出了一个框架SimCLR,通过最大化同一数据的不同数据增强处理后的两个视角之间的相似度来学习有效表征。

1. 如图所示，本文首先将数据$x$
2. $f(\bullet )$
3. 最后编码器输出的结果通过非线性变换$g(\bullet )$

3. 代码

这里仅提供文章提到的两个点的代码：

1. 数据增强
   高斯模糊

import numpy as np
import torch
from torch import nn
from torchvision.transforms import transforms

np.random.seed(0)


class GaussianBlur(object):
    """blur a single image on CPU"""
    def __init__(self, kernel_size):
        radias = kernel_size // 2
        kernel_size = radias * 2 + 1
        self.blur_h = nn.Conv2d(3, 3, kernel_size=(kernel_size, 1),
                                stride=1, padding=0, bias=False, groups=3)
        self.blur_v = nn.Conv2d(3, 3, kernel_size=(1, kernel_size),
                                stride=1, padding=0, bias=False, groups=3)
        self.k = kernel_size
        self.r = radias

        self.blur = nn.Sequential(
            nn.ReflectionPad2d(radias),
            self.blur_h,
            self.blur_v
        )

        self.pil_to_tensor = transforms.ToTensor()
        self.tensor_to_pil = transforms.ToPILImage()

    def __call__(self, img):
        img = self.pil_to_tensor(img).unsqueeze(0)

        sigma = np.random.uniform(0.1, 2.0)
        x = np.arange(-self.r, self.r + 1)
        x = np.exp(-np.power(x, 2) / (2 * sigma * sigma))
        x = x / x.sum()
        x = torch.from_numpy(x).view(1, -1).repeat(3, 1)

        self.blur_h.weight.data.copy_(x.view(3, 1, self.k, 1))
        self.blur_v.weight.data.copy_(x.view(3, 1, 1, self.k))

        with torch.no_grad():
            img = self.blur(img)
            img = img.squeeze()

        img = self.tensor_to_pil(img)

        return img

组合各类增强手段

class ContrastiveLearningDataset:
    def __init__(self, root_folder=r"D:pyprojectrepresentation_learningdata"):
        self.root_folder = root_folder

    @staticmethod
    def get_simclr_pipeline_transform(size, s=1):
        """Return a set of data augmentation transformations as described in the SimCLR paper."""
        color_jitter = transforms.ColorJitter(0.8 * s, 0.8 * s, 0.8 * s, 0.2 * s)
        data_transforms = transforms.Compose([transforms.RandomResizedCrop(size=size),
                                              transforms.RandomHorizontalFlip(),
                                              transforms.RandomApply([color_jitter], p=0.8),
                                              transforms.RandomGrayscale(p=0.2),
                                              GaussianBlur(kernel_size=int(0.1 * size)),
                                              transforms.ToTensor()])
        return data_transforms

    def get_dataset(self, name, n_views):
        valid_datasets = {'cifar10': lambda: datasets.CIFAR10(self.root_folder, train=True,
                                                              transform=ContrastiveLearningViewGenerator(
                                                                  self.get_simclr_pipeline_transform(32),
                                                                  n_views),
                                                              download=True),

                          'stl10': lambda: datasets.STL10(self.root_folder, split='unlabeled',
                                                          transform=ContrastiveLearningViewGenerator(
                                                              self.get_simclr_pipeline_transform(96),
                                                              n_views),
                                                          download=True)}

        try:
            dataset_fn = valid_datasets[name]
        except KeyError:
            raise InvalidDatasetSelection()
        else:
            return dataset_fn()

2. 非线性变换

class ResNetSimCLR(nn.Module):

    def __init__(self, base_model, out_dim):
        super(ResNetSimCLR, self).__init__()
        self.resnet_dict = {"resnet18": models.resnet18(pretrained=False, num_classes=out_dim),
                            "resnet50": models.resnet50(pretrained=False, num_classes=out_dim)}

        self.backbone = self._get_basemodel(base_model)
        dim_mlp = self.backbone.fc.in_features

        # add mlp projection head
        # 修改resnet最后一层的全连接层即可
        self.backbone.fc = nn.Sequential(nn.Linear(dim_mlp, dim_mlp), nn.ReLU(), self.backbone.fc)

    def _get_basemodel(self, model_name):
        try:
            model = self.resnet_dict[model_name]
        except KeyError:
            raise InvalidBackboneError(
                "Invalid backbone architecture. Check the config file and pass one of: resnet18 or resnet50")
        else:
            return model

    def forward(self, x):
        return self.backbone(x)

服务器托管，北京服务器托管，服务器租用 http://www.fwqtg.net