clip:learning transferable visual models from natural language supervision

CLIP 论文逐段精读【论文精读】 - 知乎知乎，中文互联网高质量的问答社区和创作者聚集的原创内容平台，于 2011 年 1 月正式上线，以「让人们更好的分享知识、经验和见解，找到自己的解答」为品牌使命。知乎凭借认真、专业、友善的社区氛围、独特的产品机制以及结构化和易获得的优质内容，聚集了中文互联网科技、商业、影视、时尚、文化等领域最具创造力的人群，已成为综合性、全品类、在诸多领域具有关键影响力的知识分享社区和创作者聚集的原创内容平台，建立起了以社区驱动的内容变现商业模式。https://www.zhihu.com/zvideo/1475706654562299904对比学习（Contrastive Learning）:研究进展精要 - 知乎对比学习(Contrastive Learning)最近一年比较火，各路大神比如Hinton、Yann LeCun、Kaiming He及一流研究机构比如Facebook、Google、DeepMind，都投入其中并快速提出各种改进模型：Moco系列、SimCLR系列、BYOL、Sw…https://zhuanlan.zhihu.com/p/367290573

自监督这块核心还是在如何构造自监督任务，对比学习是其中之一，在直白点就是如何构造正负样本对，度量学习中，毕竟有监督信号，因此正负样本对的构造旨在让相似样本接近，让不相近样本远离，对应在超球面是相似样本聚集，在自监督中，目前主流的构造带隐形标签的方式是通过数据增强，通过对原始样本进行不同方式的数据增强产生多个自增强的集合，在一个batch中，经过数据增强产生的样本本质是一个x的不同数据空间的插值，其实还是代表了样本本身的特性，因此是一个正例，一个batch中不同样本则为负例，这是单独从图像角度出发的图像自监督，图像自监督有两个方向，一个是生成式自监督（GAN/VAE），一个是判别式自监督，对比学习就属于判别式自监督，mae这种带mask的就属于前者是像素级的重构。clip属于图文架构，模式上仍然是对比学习，同一对图像和文本是一个正例，否则为负例，在clip的输入上体现就是对齐文本和图像的维度之后，在对角线上的都是正例，在非对角线上是负例，通过一个矩阵就可以构造。

1.introduction and movtiation work

        clip是openai的，openai是gpt为主的公司，说白了就是偏向生成式模型的公司，gpt是生成模型，所以基本都倾向于自回归的思路，比如mask的transformer。此外clip说加大模型加大数据量，可以在ConVIRT上有很好的的效果。

上面这张图基本就是核心图了，第一块是预训练左边的框架图，对角线是标签，是正例，其余位置是负例，右边是面对zero-shot构建的promot工程，这块用代码对着看更清晰，

2.approach

       作者在图像侧尝试了resnet，efficientnet和vit，最终选择了vit。从头开始训练clip，不借助imagent预训练的图像编码器和预先训练的权重初始化文本编码器。最后，控制softmax中logits范围的温度参数在训练期间直接优化为对数参数化乘法标量，以避免变成超参数。

3.training

        一般视觉vit，文本bert，我看到chinese_clip中，文本用的是roBERTa-base/large，图像用的vit-B-16/32，vit-L-14，其实一开始openai只有英文的clip，因此也有不少团队通过蒸馏等一些手段去对齐中文bert。一共训练了32个epoch，batch为32768。

        clip文本侧和图像侧输出的特征维度是512，dalle图像维度特征是1024，文本维度是256。

CLIP(
  (visual): VisualTransformer(
    (conv1): Conv2d(3, 768, kernel_size=(16, 16), stride=(16, 16), bias=False)
    (ln_pre): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
    (transformer): Transformer(
      (resblocks): Sequential(
        (0): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (1): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (2): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (3): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (4): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (5): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (6): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (7): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (8): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (9): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (10): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
        (11): ResidualAttentionBlock(
          (attn): MultiheadAttention(
            (out_proj): _LinearWithBias(in_features=768, out_features=768, bias=True)
          )
          (ln_1): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
          (mlp): Sequential(
            (c_fc): Linear(in_features=768, out_features=3072, bias=True)
            (gelu): QuickGELU()
            (c_proj): Linear(in_features=3072, out_features=768, bias=True)
          )
          (ln_2): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
        )
      )
    )
    (ln_post): LayerNorm((768,), eps=1e-05, elementwise_affine=True)
  )
  (bert): BertModel(
    (embeddings): BertEmbeddings(
      (word_embeddings): Embedding(21128, 768, padding_idx=0)
      (position_embeddings): Embedding(512, 768)
      (token_type_embeddings): Embedding(2, 768)
      (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
      (dropout): Dropout(p=0.1, inplace=False)
    )
    (encoder): BertEncoder(
      (layer): ModuleList(
        (0): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (1): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (2): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (3): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (4): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (5): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (6): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (7): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (8): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (9): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (10): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
        (11): BertLayer(
          (attention): BertAttention(
            (self): BertSelfAttention(
              (query): Linear(in_features=768, out_features=768, bias=True)
              (key): Linear(in_features=768, out_features=768, bias=True)
              (value): Linear(in_features=768, out_features=768, bias=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
            (output): BertSelfOutput(
              (dense): Linear(in_features=768, out_features=768, bias=True)
              (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
              (dropout): Dropout(p=0.1, inplace=False)
            )
          )
          (intermediate): BertIntermediate(
            (dense): Linear(in_features=768, out_features=3072, bias=True)
          )
          (output): BertOutput(
            (dense): Linear(in_features=3072, out_features=768, bias=True)
            (LayerNorm): LayerNorm((768,), eps=1e-12, elementwise_affine=True)
            (dropout): Dropout(p=0.1, inplace=False)
          )
        )
      )
    )
    (pooler): BertPooler(
      (dense): Linear(in_features=768, out_features=768, bias=True)
      (activation): Tanh()
    )
  )
)

4.中文clip

目前可以使用阿里巴巴开源的chinese_clip，有开源的权重和训练代码，使用大规模中文数据训练大约有2亿对里面也有悟空数据集，悟空数据集本身大约有一亿对左右，所有这个预训练还是很大的。此外，还可以使用下面这个项目转换过来的悟空数据集上训练的clip模型，这个clip是完全和openai的clip是对齐的，权重可以直接加载到openai的clip中跑。https://github.com/ELFEamp/Chinese-clip-pretrained-on-wukong