In two days, the National Day and Mid-Autumn Festival will be the long holiday . Maybe everyone's mind is ready to take off and start to soar.
Here, Da Bai first wishes everyone a happy holiday and happy every day. If time permits, often go home and have a look.
As of this week, the < Dabai AI Weekly Report > has been organized for 12 weeks. The weekly report was originally prepared for the members of the company group to understand the weekly industry trends.
Later, Dabai synchronized the content to Zhihu, CSDN, public account and other platforms, and can share it with thousands of students every week. In the interval of busy work, it can help everyone quickly and efficiently understand the industry dynamics and dig out what they need, which is also the driving force for great efforts .
If you want to view the previous weekly articles, click the link to view it.
At the same time, Dabai will also extract more systematic knowledge points for each project in all weekly reports, sort them into "Summary of Dabai AI Weekly Essence" , click to view .
In addition, Dabai is also constantly collecting and updating the video sharing of the authors and working experience of each project algorithm, hoping to improve the efficiency of some explorations , click to view .
Finishing summary: Jiang Dabai
Content cycle: 2020.09.21-2020.09.28 Synchronous public account
: Jiang Dabai
table of Contents
1 Organize the list of public accounts involved
(1) I love computer vision
(2) Cver
(3) Datawhale
(4) AI algorithm and image processing
(5
) Opencv Academy (6) PaperWeekly
(7) Xinzhiyuan (8) Jishi
Platform
(9) Machine Heart
(10) Machine Learning Algorithm Engineer
(11) AI Research Institute
(12) GiantPandaCV
(13) AI Deep Learning Sight
(14) July Online Lab
(15) Artificial Intelligence Frontier Lecture
(16) AI Technology Review
(17) Machine Learning Algorithm and Python refinement
(18) Microsoft researcher AI headline
(19) VALSE
(20) AI algorithm training camp
(21) There are three AI
(22) AlWalker
(23) Qubit
(24) AI park
(25) AI artificial intelligence beginner
(26) The road to computer vision
(27) Xiaobai learns vision
2 Summary of industry highlights
2.1 Basic knowledge
2.1.1 Deep learning related
① Network structure
(1) Attention mechanism in CV: simple and effective CBAM module
Link: https://mp.weixin.qq.com/s/gEcIcS5BSArFBX1CrPhVkA
(2) "Overfitting" can also be used as waste: someone Use it to reconstruct the surface of 3D objects in high definition, with 99% reduction in parameters
Link: https://mp.weixin.qq.com/s/otmUN_5qqpzv8M0Wboe2AA
② Model pruning & distillation & acceleration
(1) Distillation technology in neural network, starting from Softmax
Link: https://mp.weixin.qq.com/s/iwXV98gkl2Ql6Mw-xlqLaQ
(2) In-depth understanding of NNIE quantitative technology
Link: https://mp.weixin.qq.com/s/giCQknGsoodNE0azbLCW0A
2.2 Research direction
2.2.1 Target detection
(1) OpenImage champion scheme: Use separate feature maps for classification and regression tasks in object detection
Link: https://mp.weixin.qq.com/s/23khMFKzYpiRZD12MtaXxg
(2) Take off | Apply YOLOV4-DeepSort to achieve the goal Tracking
link: https://mp.weixin.qq.com/s/POjff6TRVBCkQ0MEpCoKFA
(3) BBAVectors: An Anchor Free rotating object detection method
Link: https://mp.weixin.qq.com/s/VpM1xY6Y55X7CAaJpJ6EwA
( 4) Loss inventory of target detection loss function
Link: https://mp.weixin.qq.com/s/oGe0so06KwLWtrxqzPtAGA
(5) Overview: Multi-scale detection methods in target detection
Link: https://mp.weixin.qq .com/s/HW0SNf-NYjXqJwlungRFpg
(6) Summary: Solve the problem of sample imbalance in target detection
Link: https://mp.weixin.qq.com/s/iOAICJege2b0pCVxPkvNiA
2.2.2 Significant target detection
(1) It is not enough to cut out the picture as fine as the hair. The new Adobe method can process 6000×6000 high-resolution images.
Link: https://mp.weixin.qq.com/s/vzFF9B0A1LTD3GcQQeWSlg
2.2.3 Face recognition
(1) The latest progress in face recognition-a summary of several related papers
Link: https://mp.weixin.qq.com/s/3TZ-1fU190y8Dr16jMoUFw
2.2.4 OCR
(1) Some new attempts at handwriting and scene text analysis and recognition
Link: https://mp.weixin.qq.com/s/LhSU19wKWuq2gBnr1j_KCw
(2) A further 64% reduction, 3.5M ultra-lightweight Chinese and English OCR model open source, The effect of stature reduction is improved.
Link: https://mp.weixin.qq.com/s/QpLcEV44MPj9Nj1KPaCrxA
(3) 3.9MB ultra-small and super strong text recognition model, supporting recognition of 20,000 characters, Ping An Property & Casualty Insurance proposes Hamming OCR
link: https://mp.weixin.qq.com/s/bNlU2GV8yL_D3wTkkLOTbw
2.3 Project application
(1) 100-year-old photo restoration algorithm | Taobao sells for $50, now teaches you for free
Link: https://mp.weixin.qq.com/s/DXSK6uHH-EEvec-M61ahsw
2.4 Opencv aspect
(1) Practical image processing operations based on OpenCV
Link: https://mp.weixin.qq.com/s/SnK3hO_Q6xazCqHBR-TlUg
3 Summary of research & industry trends
3.1 Research related
(1) C++ realizes neural network from scratch (collector's edition: 20,000 words long)
link: https://mp.weixin.qq.com/s/oLhyxa0e_KVygwI8vHUtdw
(2) I have to praise it! A (probably) the best website for acquiring massive CV datasets in China.
Link: https://mp.weixin.qq.com/s/tcrRiDoyVsEfi9aPDaUvJA
(3) This week’s excellent open source project sharing: free mobile real-time face 212 key points 7 major projects including SDK and nlpcda Chinese data enhancement package
Link: https://mp.weixin.qq.com/s/utP7LBWJ7emAnj0i6XtnJw
3.2 Industry News
(1) Foreign Handmade Geng | "Hundred shots and 100 hits" rebounds have been updated!
Link: https://mp.weixin.qq.com/s/GjNudL-Wep-43vdrYBeULg
(2) Shen Xiangyang: Talking about artificial intelligence creation
Link: https://mp.weixin.qq.com/s/BUJ5O5xlOLWA_SK4Hg35sA
( 3) It turns out that the strongest son-in-law is long in the second dimension, and the AI special effect that penetrates the second dimension wall is double-fired.
Link: https://mp.weixin.qq.com/s/4Hsunit4USqegMrBxc4Juw
(4) Some suggestions: for the year I am just a programmer
Link: https://mp.weixin.qq.com/s/ZSVIRGiVwKQ_HvP3tHufvQ
4 Pytorch basic content
(1) PyTorch's Autograd detailed explanation
Link: https://mp.weixin.qq.com/s/sZrmk18SXortVLX8dwiksQ
5 Algorithm & operating system knowledge
(1) What are the methods for image noise reduction?
Link: https://mp.weixin.qq.com/s/aapDAsPkrVsYulMOOx3LKg
(2) A quick reference manual for common commands of Linux system, Xiaobai strongly recommends collecting!
Link: https://mp.weixin.qq.com/s/wKuKZHwWqQcWVdJ1VdylAw
6 Summary of Top Conference Highlights
(1) DCTNet
link for CVPR 2020 to learn in the frequency domain : https://mp.weixin.qq.com/s/3-NEbE6aZHR6dgW6GCmldw
(2) ECCV 2020 paper inventory-personnel re-identification (ReID)
link: https ://mp.weixin.qq.com/s/pnlldquTloTyea2j7dA83A
(3) ECCV 2020 paper inventory-salient target detection
link: https://mp.weixin.qq.com/s/oWcPutsRD0bOanrvpwnCpw
(4) ECCV 2020 paper Large inventory-3D target detection
link: https://mp.weixin.qq.com/s/BTKKi_1mk6AbmO6xTVANCw
(5) ECCV 2020 paper inventory-target detection
link: https://mp.weixin.qq.com/ s/dvue9Kx6UQ5iGE-Z615Tug
(6) ECCV 2020 Paper Inventory-Face Technology
Link: https://mp.weixin.qq.com/s/7BphCe8YYW1_13jWDVYvpw
