注：此DINO并非自蒸馏自监督的那个[[DINO]]，而是派生自

Text Encoder

使用的backbone是BERT(通过MLM训练)
该研究认为，image encoder的模型大小应该大于text encoder,所以在text encoder这里，只使用六层self attention来提取特征，剩余六层cross attention用于multi-modal encoder。

ITC Loss & Momentum

参考Moco [[Moco- Momentum Contrast for Unsupervised Visual Representation Learning]]

Improve Noisy Web Data

见[[BLIP]]，是沿用的工作

Loss

ITM

MLM

,

通过结合[[DINO]]和grounded-pretraining，可以使用人类输入（例如类别名称或转介表达式）检测任意对象
Open-Vocab. Det

an open-set object detector that can detect any objects with respect to an arbitrary free-form text prompt. The model was trained on over 10 million images, including detection data, visual grounding data, and image-text pairs. It has a strong zero-shot detection performance. However, the model needs text as inputs and can only detect boxes with corresponding phrases.

Grounding-DINO

Principle

Tight modality fusion based on [[DINO]]

什么是feature fusion?

https://blog.csdn.net/h661975/article/details/135116957

loss: ITC (Image Text Contrastive)

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24

# image_encoder - ResNet or Vision Transformer 
# text_encoder - CBOW or Text Transformer 
# I[n, h, w, c] - minibatch of aligned images 
# T[n, l] - minibatch of aligned texts 
# W_i[d_i, d_e] - learned proj of image to embed 
# W_t[d_t, d_e] - learned proj of text to embed 
# t - learned temperature parameter  

# extract feature representations of each modality 
I_f = image_encoder(I) #[n, d_i] 
T_f = text_encoder(T) #[n, d_t]  

# joint multimodal embedding [n, d_e] 
I_e = l2_normalize(np.dot(I_f, W_i), axis=1) T
_e = l2_normalize(np.dot(T_f, W_t), axis=1)  

# scaled pairwise cosine similarities [n, n] 
logits = np.dot(I_e, T_e.T) * np.exp(t)  

# symmetric loss function 
labels = np.arange(n) 
loss_i = cross_entropy_loss(logits, labels, axis=0) 
loss_t = cross_entropy_loss(logits, labels, axis=1) 
loss = (loss_i + loss_t)/2

Cross_entropy_loss:

CLIP 本质上是全局图像嵌入，不利于像素对齐特征提取。