"Vision Transformer Punch Camp" is officially launched! Dr. Zhu Yi, a senior researcher in the Deep Learning Laboratory of Baidu Research Institute , gave lectures throughout the course, and six professional teaching assistants answered questions throughout the course. Ten lessons will show you the ViT model! Taught in Chinese, completely open source!
Class start time: November 23rd - December 2nd, live lectures will start on time at 20:30 every night~
Course link (playback, supporting practice): https://aistudio.baidu.com/aistudio/course/introduce/25102
Official Q group (answering questions, technical exchanges): search the group number 920200490 or scan the QR code below to reply to the keywords to get the way to join the group~
Since Transformer was proposed in 2017, it has gone from sweeping the NLP field to being questioned about its effectiveness in CV tasks, and not long ago, it has shown excellent performance that is close to CNN in many image tasks. And ICCV2021 best paper: Swin Transformer is released, Transformer gradually shines in the CV field!
Therefore, words such as Transformer, Vision Transformer, QKV, and Attention is all you need once again detonated the academic circle!
As a developer, while applauding this academic superstar , why not get on the bus in time to figure out what the QKV in the encoder and decoder is, and what the popular Swin Transformer has done right? Why can Transformer sweep NLP , and how does it work in CV tasks? How does the MAE algorithm work in the latest CV masterpiece ?
Spend 10 hours to really understand the technical value behind these hot words , implement these popular ViT algorithms by yourself , and even apply them to your current work and scientific research to become the next top hit!
"ViT Ten Lectures" Is All You Need!
On November 23 (Tuesday), Dr. Zhu Yi, a reviewer of AI Summit and a senior researcher of Baidu Deep Learning Research Institute, gave a live lecture of "Ten Lectures on Vision Transformer". 1 hour per night for 10 consecutive days , Dr. Zhu Yi will take you to play the ViT algorithm from scratch!
Vision Transformer's cutting-edge technologies are comprehensively explained: from ViT classic algorithms to the latest academic frontiers, from technical principles to implementation details, ViT, Swin, and DETR will take you one by one to master the new paradigm of vision algorithms.
Thesis analysis + line-by-line Coding, on-site writing on the blackboard, hand-pushing formulas: Detailed analysis of the main points of the thesis, half of each class will lead you to write code on the spot ~ show you your own ViT model, whichis also very friendly to Xiaobai .
The whole process of carrying and playing with ViT: the course content is completely open source , the whole course is taught live in Chinese , the high- level teaching assistant team answers questions in the community, and there are practical tasks of different difficulty to meet the needs of each stage of learning, and free GPU computing power support is provided!
Join the open source technology team, and become the core contributor of the academic suite PPViT together with the leaders of the research institute: jointly lower the threshold for the realization of the ViT model and improve the productivity of scientific research and learning. (https://github.com/BR-IDL/Paddle ViT )
you will gain
Whether you are new to deep learning or are already doing scientific research, whether you want to switch to NLP from CV, or want to engage in CV from NLP, or you want to use the latest visual technology to play competitions and publish papers, by learning "Vision Transformer Ten Lectures" ", you will gain the ability to turn the model diagram in the paper into a line of code, say goodbye to simple git clone and package adjustment!
Instructors
Dr. Zhu Yi, senior researcher of the Deep Learning Laboratory of Baidu Research Institute, Fei Paddle PPDE (Technical Expert for Developers), and Reviewer of AI Top Conference. With front-line work experience in Silicon Valley, USA, he is currently in charge of deep learning and visual algorithm research and development at Baidu Research Institute. He has published many high-level papers. In recent years, he has won the top ranking in international competitions such as CVPR, ICCV, ECCV and other top conferences.
Course Outline
Lecture 1
Theory: Hello, Vision Transformer!
Practice: Warmup: Model building and training
Lecture 2
Theory: From Transformer to Vision Transformer
Practice: Play with Tensor operations and start building ViT Lecture
3
Theory: You see yours, I see My Detailed Attention
Practice: Multi-Head Self Attention
Lecture 4
Theory: Build your first ViT model
Practice: How to implement the ViT model
Lecture 5
Theory: The ViT model is built, how to train it efficiently?
Practice: actual combat model building and training
Lecture 6
Theory: What is Window Attention?
Practice: Attention Mechanism on Image Window
Lecture 7
Theory: The famous Swin Transformer
Practice: Implementing your second ViT model
Lecture 8
Theory: Combination of Conv and Transformer
Practice: How to implement data loading from the framework source code
Ninth Lecture
Theory: Take you to understand cutting-edge algorithms: Visual BERT
practice: Model training skills
Lecture 10
Theory: New paradigm of detection algorithm-DETR
practice: The whole process of actual ViT training and testing
Wan Hao Li, just waiting for you
Participate in the course and pay tribute to the open source , and you can get: PaddlePaddle official certificate of completion , Marshall ACTON II BLUETOOTH wireless subwoofer, HHKB Professional electrostatic capacitance Bluetooth keyboard, Tesla V100 GPU computing card, Baidu network disk super membership and many other rewards! Not capped! There is also a fancy lottery in the live room waiting for you to get it!
write at the end
"If you take it from the top, you can get it from it; if you take it from the top, you can get it from the bottom; if you take it from the bottom, you won't get anything." I locked the ViT punch-in camp and hoped that everyone would broaden their horizons, set high goals, and finally achieve their own satisfaction. the result of!