June 13-19, 2022 AI industry weekly (issue 102): career development

Chapter one ： In the hospital

I spent several days with my wife in the hospital last Thursday , Because in late pregnancy , Various indicators need to be checked .

When I first moved in, because of the same room , I met two other villagers , The back is vacant because of other rooms , Three people are divided into three rooms .

Because I get along well , All kinds of food in the next few days 、 Fruits 、 Ice cream sundae , Several families send each other .

I suddenly remembered my childhood , There are no high-rise buildings now 、 There are no cars or electric cars , But at that time, my childhood was barefoot , In the happiness of walking through the streets .

Every household , Next door neighbor , Yes, after dinner , Drop in at random , Leaning on the door chatting .

Now there are many kinds of electronic products , But the child's childhood , But I don't have the feeling of childhood .

Upstairs Downstairs , Even the doors of neighbors are closed , Children's childhood , They are also occupied by various training courses , Afraid of being left behind by other children .

Of course , Being a parent , I have reached the age of being old and young , working pressure , Family pressure , YedU
It is not easy to , This may also be the current situation of every Chinese family .

Chapter 2 ： Occupation development

There was a friend last Friday , Ask me on wechat ： I think it's AI Technology is to take the route of deep excavation , It is better to take the breadth route ？

I believe that every friend who has worked for more than two years , Will encounter such confusion .

from 19 I have been engaged in AI Industry start , Dabai's career path , From algorithmic technology -> The algorithm is responsible for ->AI The overall responsible route of the plate , It's a route that tends to be broad .

Of course, there are also many skilled people around , Deep excavation AI technology , Self research framework , Self developed algorithm , Ranking of various technical competitions , Dabai can only sigh that he is not as good as .

But if which one is better , In fact, it still varies from person to person , Both have made good progress , Of course, both of them face different pressures .

In terms of development , There are many skilled people around who earn more than one million dollars , Their universal qualities , Have a good academic background , Good company , At the same time, I am very proficient in a certain piece of technology , Have some knowledge of related technologies .

Of course , Technical pressure , Relatively pure , At work , It mainly solves many technical bottlenecks .

And from a wide range of routes , As far as I am concerned , The current pressure , Existing technical pressures , There are also teams 、 Pressure in product direction .

In terms of technology, , Polish a product , Need full stack technology , Then you have to find a way to find these team members , With talents, there is hope .

Remember the previous leaders of the company , When I first started , Once said a word to me ： You must have a team to develop .

At that time, it was mainly technical , No special feelings , Focus on your own work , Now I feel deeply .

And in terms of direction ,AI Technology is the base of products , Find the market 、 Find opportunities 、 Find a team , Grinding products , It is the responsibility of the person in charge .

After being hospitalized with his wife last Thursday night , Friday to Saturday night 12 spot , I have been revising in the delivery room AI Of the product PPT, And on Sunday morning VP、 All directors meet together to discuss , It's also very stressful .

So every aspect has a good side , There is also the pressure side , Of course , Many people always see the good side , And ignore the back side .

And back to the beginning , It is better to take the depth route or the breadth route ？

Dabai thinks , Or according to their own character , Interest in , And preferences .

But when I first started , You can give yourself two years , Think hard and try , Give yourself more choices in two years .

Dabai has always felt , Challenge yourself to do the right things that you find difficult , Don't stay in the comfort zone , The road of life can always go up .

Previous weekly report , The great white society collages the essence of the week. , Put it in order 《 Great white AI Weekly Digest 》 in , Click to view .

What kind of project knowledge do you need in the later stage , You can go straight to Corresponding reading .

Dabai is there, too Continuously collect updates Video sharing of algorithm authors and professional experience of various projects , I hope you can improve Efficiency of exploration , Click to see .

Organize and summarize ： Jiang Dabai
Content cycle ：2022.6.13-2022.6.19
Synchronization official account ： Jiang Dabai

---

1 Sorting out official account list

（1） I love computer vision
（2）Cver
（3）Datawhale
（4） qubits
（5） Polar market platform
（6） The new intellectual yuan
（7） Almost Human
（8）AI Algorithm and image processing
（9）Opencv School
（10）PaperWeekly
（11） Machine learning algorithm engineer
（12）AI Workshop
（13）GiantPandaCV
（14）AI Deep learning vision
（15） Online lab in July
（16） Lectures on the frontiers of artificial intelligence
（17）AI Technology Review
（18） Machine learning algorithm and Python Intensive research
（19）AIZOO
（20） Microsoft researcher AI headlines
（21）VALSE
（22）AI Algorithm training camp
（23） There are three AI
（24）AlWalker
（25）AI park
（26）AI AI beginners
（27） The way of computer vision
（28） Xiaobai studies vision
（29）HyperAI Supernerve
（30） Intellectual book boy
（31） Computer vision life
（32）CV circle

2 Summary of industry essays

2.1 Basic knowledge

2.1.1 Deep learning is related to

（1） Label regularity ： Label smoothing 、 Label distillation and self correction are similar
link ：https://mp.weixin.qq.com/s/W9f7XF9PWEEC1zi7jDM0vA
（2） How to split data sets correctly ？ Three common methods are summarized
link ：https://mp.weixin.qq.com/s/p0CY-A5oJ4lfcj2f8MIgdQ
（3）Adam That's great , Why are you still right SGD Not forget for a moment ？ Understand the depth of learning optimization algorithm
link ：https://mp.weixin.qq.com/s/a-FqHwJcfbPQMBMj668pNw
（4） Reexamine pooling ： Your feeling field is not ideal
link ：https://mp.weixin.qq.com/s/ndLKNSDJ9_9B-J4d0_yD4w

2.2 Research direction

2.2.1 object detection

（1）labelGo： be based on YOLOv5 Reverse annotation software , Improve labeling efficiency
link ：https://mp.weixin.qq.com/s/5Y-Hqk2hyaDQJwmF3VB5Rg
（2）YOLOS： Rethink through target detection Transformer（ With source code ）
link ：https://mp.weixin.qq.com/s/89Se8yFQBiF4s1BKl1-qGA
（3）CVPR2022 | iFS-RCNN： An incremental small sample instance divider
link ：https://mp.weixin.qq.com/s/Wn6iyBwGMVYJBQZx6ceDdA

2.2.2 Counter samples

（1） Transferable attention attack （TAA）
link ：https://mp.weixin.qq.com/s/t9z2SEA1ZhXE86l6XjORuw

2.2.3 3D Point cloud

（1） Reinvigorate PointNet++ Strong wind ！PointNeXt： Reexamine through improved model training and scaling strategies PointNet++
link ：https://mp.weixin.qq.com/s/p2ZFxUIHLfjSWrHy9nkzMQ

2.2.4 Face recognition

（1） Face shadow suppression , Illumination robust face recognition algorithm
link ：https://mp.weixin.qq.com/s/Yl3Wqvqu4v03Q3iGJCf2sw

2.2.5 Lightweight network

（1） Lightweight network MobileNet A new paradigm of pruning compression , It is worth learning from ！
link ：https://mp.weixin.qq.com/s/rmVhIu94zKd7eem3DC16pw

2.2.6 Image enhancement

（1） Fast 、 Flexible and robust low illumination image enhancement methods （​CVPR 2022 Oral）
link ：https://mp.weixin.qq.com/s/8PNadR8wyKYvATAKLIPa9Q

2.3 Project practice

（1） Use Python+OpenCV+Keras The penalty generation of vehicle driver without mask based on license plate is realized
link ：https://mp.weixin.qq.com/s/emuqorqDqLL8kYpMW88cZw
（2） be based on YOLOv5 and U-NET Fire detection and segmentation
link ：https://mp.weixin.qq.com/s/kx_YjUL-2d5XoN6BO7uxZA
（3）[ Open source project ] be based on FPGA Video image mosaic and fusion
link ：https://mp.weixin.qq.com/s/ycvWFnyrXzmDj52xD46vOA

3 industry & Expand reading dynamics

3.1 Industry trends

（1） tsinghua &BAAI Tang Jie's team proposed the first open source universal large-scale pre training text to video generation model CogVideo, contain 94 100 million large number of parameters ！ The code is going to be open source ！
link ：https://mp.weixin.qq.com/s/Um91aUgHXUeIg5PiezBqsg
（2）Apple new MobileOne： The mobile terminal only needs 1ms High performance backbone ！ transcend MobileViT！
link ：https://mp.weixin.qq.com/s/3wA6QJ2anViwDEOd9h-h4A
（3） give up TensorFlow, Google is turning to JAX
link ：https://mp.weixin.qq.com/s/UMIJltvUWfvf8USub5WetQ
（4） Dr. SunJian, the chief scientist of Kuangshi, passed away , How to evaluate his contribution ？
link ：https://mp.weixin.qq.com/s/Yg7KMOXNIep-4juyshYlDw
（5）49.3% Of the researchers said “ Have to spend a lot of time and energy applying for funds ”！ How to make scientific researchers sit still “ cold bench ”？
link ：https://mp.weixin.qq.com/s/TPK31gVVBJpq_0NOmi5zFQ

3.2 Expanding reading

（1） The pictures drawn by digital people in tens of seconds have been sold 17 ten thousand , The real artist envied and cried
link ：https://mp.weixin.qq.com/s/0SV8KPeOtYdsoh7ZPiMSkA
（2） finally ,Yann LeCun Send a document to refute Gary Marcus： Don't regard temporary difficulties as hitting the wall
link ：https://mp.weixin.qq.com/s/-Z6VrxXQwFHF-b1eSJf0Ig
（3） Shi Yigong ： Until I graduated from my doctor's degree , Not interested in research ！ Confused about the future , I don't know what to do in the future …
link ：https://mp.weixin.qq.com/s/QZNCixkyUwwTMPdoNBFfRQ
（4） Grieved ！ Another national outstanding youth died of illness , They have left one after another within a week 4 A young and middle-aged scholar
link ：https://mp.weixin.qq.com/s/LvQHydAG5Kcxnw5TqA5gTg

4 Basic knowledge of

（1） Tai Chi graphics 60 Lines of code to achieve the classic thesis ,0.7 Second Poisson disk sampling , Than Numpy Fast implementation 100 times
link ：https://mp.weixin.qq.com/s/eLFttlQR5NxxYY7O8ONCMg
（2） Some knowledge points of image convolution and filtering
link ：https://mp.weixin.qq.com/s/yvDfpcWXiE3yUu6NisGeMA

5 Martial arts script

（1） The most comprehensive in-depth learning surface in history has been summarized （ A detailed explanation of the answer is attached ）
link ：https://mp.weixin.qq.com/s/6ihwQA_t0Qa3aBZ35CL_Hw
（2） Martial arts script | vivo You must see the secret script when you apply for an algorithm post
link ：https://mp.weixin.qq.com/s/qFX-23LC-8XNHuyi_cDWXQ

6 Data sets

（1） The largest human image matting data set at present P3M-10k Open source ！ Image matting research that helps protect privacy
link ：https://mp.weixin.qq.com/s/AzSaMxvtEn_V17axQhmzIw