Why can transformer break into the CV world and kill CNN?

【CV Transformer Latest progress 】2021 year 8 month , Zurich Federal Institute of Technology ETH Put forward Swin Transformer The algorithm is applied in image restoration , In the classic image 、 Real scene image super segmentation 、 Image noise reduction and other fields have achieved very good results .

From the date of submission ,Transformer The model is already in CV、NLP And more 「 Show off 」, Strength impact CNN.

Transformer Why are you so powerful ？ Because it's classifying 、 It shows extremely strong performance in detection and other tasks . Moreover, the development of backbone network also promotes the development of downstream tasks ,Swin Transformer It has become a Tu Bang like existence , It has broad application prospects in industry . Therefore, it has aroused the strong interest of artificial intelligence graduate students .

But to get through CV Transformer The difficulty is not small. ： One side ,Transformer This is applied to NLP The paper of , A lot of it has reached a consensus , These consensus contents will not be introduced in detail in the paper , for example QKV What is it? ,embedding What is it, etc , It's hard for people in other directions to understand .

On the other hand , Nearly half a year ,Transformer+CV My thesis already has 40 Multiple articles . How quickly academic research is updated , It is directly proportional to the speed of hair loss

If fans want to Systematic and efficient learning CV Transformer, I recommend you to join the eye of depth 【CV Transformer Live broadcast of the thesis 】

↓ Front welfare  ↓

The original price 399 element , Now? 0.1 Yuan to collect ！

Buy and give away 《 Efficiency improvement 3 Times Paper Reading methods 》

↓ Scan the QR code below to sign up immediately ↓

 

CV Expert methodology , Teach you to study papers systematically

Transformer Master takes the lead   save 21 Days of thesis study time

Deep eye electronic sheep tutor combines his own work and learning experience , And cooperate with the grinding of the deep eye teaching and research group , Sum up a CV Transformer The learning path of ：

2 Live broadcast + Record and broadcast    Tamp  CV Transformer Basics

Step1： Systematic understanding of CV Transformer Technological evolution path and development history

Step2： Work: CV Transformer Cornerstone thesis — ViT

《An Image is Worth 16x16 Words:Transformers for Image Recognition at Scale》 abbreviation ViT.ViT yes Google stay 2020 The first article proposed in used pure transformer To carry out the task of image classification , Its value lies in showing that CV Use pure Transformer Structural possibilities , Much of the later work is based on ViT To improve .

And this model has only been released for more than half a year ,github On ViT Of repo There are many , be based on tensorflow and pytorch Both have .star The number is already several thousand , Visible influence  . Personal feeling ViT It has a great impact on subsequent papers , Many papers draw on VIT The relevant practices inside .

Deep eye electronic sheep tutor   Will be taken from   Research background   To   Algorithm model , Take you through ViT！

① Dig deep into the research background  

The outline leads , from 4 Large dimension introduction paper , Explain the research background of the paper in depth 、 Achievements and significance , Introduce the core achievements of the paper , Compare and solve the same problem , The advantages and disadvantages of the existing solutions and the new solutions proposed in the paper , Be familiar with the overall idea and framework of the paper , Establish a general understanding of this paper .

②  Dead knock algorithm model

The teacher will focus on the model principle in the paper , Deeply disassemble the model structure , Gradually deduce the key formula , Let you know how each factor of the algorithm affects the result , Master the experimental means and results , The teacher will help you pick out the key points in your paper 、 Innovation and inspiration , Save yourself time groping .

Master with learning Together to promote

· 3 High quality community service , The tutor accompanies the whole process

· 2 Live broadcast + Record and broadcast , Turn on CV Transformer  A new chapter

· 100+ Students communicate with the group , Learning experience up up

·  Assistant 24 Hour answer questions , No longer afraid of debug

· Exclusive class teacher private letter supervisor , Treat learning procrastination

Give value after class 298 Yuan learning package

In order to motivate everyone to complete their study , We have also prepared value 298 yuan Algorithm Engineer Interview Kit . Just finish all the courses , You can send a private letter to the head teacher to get ！

The students praised , It's delicious ！

This time, , I applied to my fans 30 Live broadcasts Welfare quota , The secret script will be given as soon as you join ：《 Efficiency improvement 3 Times Paper Reading methods 》.

↓ Front welfare  ↓

The original price 399 element , Now? 0.1 Yuan to collect ！

Buy and give away 《 Efficiency improvement 3 Times Paper Reading methods 》

↓ Scan the QR code below to sign up immediately ↓

CV Expert methodology , Teach you to study papers systematically

If you don't know how to read a paper 、 I don't know how to reproduce the paper correctly , Be sure to follow this course once , Because the right way can save you 10 Times the reading time .