'분류 전체보기'에 해당되는 글 1027건

  1. 2020.10.21 matplotlib으로 figure창을 여러개 띄우기
  2. 2020.10.13 안녕하세요~ 저희 그룹에서 지난 반기동안 진행된 Sensor Fusion Study 관련 전체 유튜브 및 슬라이드쉐어 스터디자료를 깃헙으로 정리
  3. 2020.10.13 안녕하세요, TensorFlow KR 여러분! StarGAN 리뷰 및 코드 실습 영상을 만들어 공유합니다. 한 장의 사진을 업로드해서 성별/나이
  4. 2020.10.12 Deep Reinforcement Learning Course (CS285) by Sergey Levine, Assistant Professor
  5. 2020.10.12 안녕하세요! 최근 모두의 말뭉치가 새로 공개된 겸하여 KoELECTRA-v3를 새로 만들었습니다:) 모두의 말뭉치를 포함한 약 34GB 데이터
  6. 2020.10.08 10 Best Machine Learning Courses in 2020
  7. 2020.10.08 Made With ML Topics A collection of the best ML tutorials, toolkits and researc
  8. 2020.10.07 Having an unbalanced dataset is not an uncommon problem in real world. Intereste
  9. 2020.09.29 Multi-View Brain HyperConnectome AutoEncoder For Brain State Classification Ban
  10. 2020.09.23 State of the art in 3D dense face alignment! For project and code/API/expert r
  11. 2020.09.18 [self][Python NumPy] 배열을 수평(옆)으로, 수직(위아래)으로 붙이기 : np.r_, np.c_
  12. 2020.09.16 인터넷에서 30종 인기 블로그나 온라인 사이트를 정리해 공유하였습니다. 참고가 되었으면 좋겠습니다~
  13. 2020.09.15 [self] PR(Precision Recall) curve VS ROC curve
  14. 2020.09.15 풀스택 딥러닝 [https://course.fullstackdeeplearning.com/](https://course.fullstackdee
  15. 2020.09.15 Top 10 Free Machine Learning Courses
  16. 2020.09.11 https://youtu.be/FeBhlDLAddc Converting video (mp4) file to audio (mp3) file usi
  17. 2020.09.09 안녕하세요! super resolution과 관련한 정말 기초적인 질문이 있어 왔습니다!ㅜ.ㅜ 이미지 분류는 resnet과 같은 CNN 기반의
  18. 2020.09.09 Bayesian Data Analysis course Vehtari et al., Aalto (CS-E5710) : https://avehtar
  19. 2020.09.02 안녕하세요, SLAM 공부중인 장형기입니다. 이번에 Visual-SLAM에서 필수적으로 사용되는 RANSAC, 그리고 에러 데이터를 (outl
  20. 2020.08.31 1주 1논문 리뷰 프로젝트 공유 안녕하세요. TFKR! 대학 여름방학 마지막 주를 맞아, 제가 이번 방학동안 진행한 1주 1논문 리뷰 프로젝트
  21. 2020.08.27 안녕하세요. 모두연 여러분. 얼마전에 열렸던 AI Grand Challenge 4차 대회의 경량화 트랙에서 사용했던 구현체를 공개합니다 (최종
  22. 2020.08.27 A novel neural network to generate high-resolution images For project and code/A
  23. 2020.08.26 안녕하세요, 왕초보입니다. MacOS, Windows 또는 LInux ? 인공지능을 제대로 공부하려는데요, 어떤 운영체제에서 시작하는 것이 성
  24. 2020.08.26 안녕하세요. ECCV 메인 컨퍼런스도 이제 이틀차에 접어 들었습니다. 어제 오프닝에서 다른 학회들과 마찬가지로 각종 시상식 코너가 있었는데,
  25. 2020.08.26 Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from
  26. 2020.08.25 Super Fast and Accurate 3D Object Detection based on LiDAR Fast training, Fast i
  27. 2020.08.25 간단히 바로 가져다 쓸 수 있는 easyOCR API 소개 드립니다. 강원도 놀러갈 때마다 보이는 간판인데, 신나기 시작하는 "인제 - 신남"
  28. 2020.08.24 SNU AI (구 SNU TF) 스터디 모임 7기에서 매주 발표되는 논문리뷰 자료들을 정리하여 TF Korea 분들께 카드뉴스형식으로 정리한 S
  29. 2020.08.24 안녕하세요. 새로 운영진에 합류하게 된, 그리고 비엔나에서 로봇 비전 연구하는 박기루 입니다. 운영진이 되고나서.. 첫글로 제 논문 홍보를 하기
  30. 2020.08.24 안녕하세요 lidar slam 공부 중인 기섭입니다. lidar slam 과 딥러닝을 어떻게 접목해야 할지 평소에 고민이 많은데요, 다름이

파이참으로 파이썬 코드 실행 중, figure을 여러개 띄울 때, 창을 닫아야만 그 다음 코드로 진행이 되었다. 이것이 불편해 찾아보니 

import matplotlib.pyplot as plt

x=[[1,2,3,4],[1,2,3,4],[1,2,3,4],[1,2,3,4]]

y=[[1,2,3,4],[1,2,3,4],[1,2,3,4],[1,2,3,4]]

for i in range(len(x)):

    plt.figure()

    plt.plot(x[i],y[i])

    # Show/save figure as desired.

    #plt.show()

# Can show all four figures at once by calling plt.show() here, outside the loop.

plt.show()

src: stackoverflow.com/questions/19189488/use-a-loop-to-plot-n-charts-python

에서처럼 그리고자 하는 그림들을 plt.figure() plt.plot(x)으로 각각 그려준 뒤, 맨 마지막에 plt.show()를 1번만 호출하면 전체창이 한꺼번에 띄어졌다.

 

import matplotlib.pyplot as plt

x=[[1,2,3,4],[1,2,3,4],[1,2,3,4],[1,2,3,4]]

y=[[1,2,3,4],[1,2,3,4],[1,2,3,4],[1,2,3,4]]

for i in range(len(x)):

    plt.figure()

    plt.plot(x[i],y[i])

    # Show/save figure as desired.

    plt.show()

# Can show all four figures at once by calling plt.show() here, outside the loop.

#plt.show()

처럼 하면 창을 매번 닫아야 그 다음이 진행된다.

 

 

 

 
 
 
 

'Python > Matplotlib' 카테고리의 다른 글

matplotlib에서 figure창을 자동으로 닫기  (0) 2020.10.21
Posted by uniqueone
,

안녕하세요~ 저희 그룹에서 지난 반기동안 진행된 Sensor Fusion Study 관련 전체 유튜브 및 슬라이드쉐어 스터디자료를 깃헙으로 정리하여 공유드립니다. 🙂 스타 많이 눌러주세요~

마지막 시간에는 실제 센서퓨전이 현실에서 어떻게 응용되는지 imu/gps/radar/lidar/camera fusion 관련 작은 세미나도 가졌습니다. 그동안 훌륭하신 스터디 원분들과 함께해서, 지식적으로 성장 뿐만아니라 좋은 에너지와 보람을 느꼈던 시간이었습니다. 함께 잘 이끌어주신 hayden님 너무 감사드려요.

저희가 학습한 KF, EKF, UKF, H infinity Filter, Particle Filter 등을 kitti dataset 을 파싱해 python 코드로 직접 구현하는 스터디를 Sensor Fusion 2 에서 진행할 예정이니 많이 기대해주세요~


Skip to content

Sign up

ai-robotics-kr/sensor_fusion_study

Sensor Fusion Study Based on "Optimal State Estimation"

 7 stars  0 forks

 Star

 Watch

Code

Issues

Pull requests

Actions

Projects

Security

Insights

Dismiss

Join GitHub today

GitHub is home to over 50 million developers working together to host and review code, manage projects, and build software together.
Sign up

 master 

Go to file

Latest commit



howtowhy



22 days ago

Git stats

Files

View code

README.md

Sensor Fusion Study

Sensor Fusion Study Based on "Optimal State Estimation - Dan Simon"

Sponsor by AI Robotics KR https://www.facebook.com/groups/airoboticskr/



Members

Study Lead :

Stella Seoyeon Yang @howtowhy, Hayden

Study Contributors :

Sora Kang, Dong Hyun Kim, Jeong Eun Park @jeongeun980906, Jin Hyuk Song @Jinhyeok1489, Young Beom Kim @eddy979797, Hae Gu Lee, Sung Min Ahn @aminsung, Soo Young Kim @kimsooyoung, Young Joo Kim, Kile

𝐂𝐡𝟏. 𝐋𝐢𝐧𝐞𝐚𝐫 𝐒𝐲𝐬𝐭𝐞𝐦 : Hayden

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch1-linear-system-hayden

Youtube : https://youtu.be/-6pO8G8PHC4

𝐂𝐡𝟐. 𝐏𝐫𝐨𝐛𝐚𝐛𝐢𝐥𝐢𝐭𝐲 𝐓𝐡𝐞𝐨𝐫𝐲 : Seo Yeon Stella Yang

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-probability-theory-stella

Youtube :https://youtu.be/38dVDay2XnA

𝐂𝐡 𝟑. 𝐋𝐞𝐚𝐬𝐭 𝐒𝐪𝐮𝐚𝐫𝐞 𝐄𝐬𝐭𝐢𝐦𝐚𝐭𝐢𝐨𝐧 : Sora Kang, Stella, Hayden

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch3-least-square-estimation-stella-hayden

Youtube : https://youtu.be/C_gtrXk3zNo

𝐂𝐡𝟒. 𝐏𝐫𝐨𝐩𝐚𝐠𝐚𝐭𝐢𝐨𝐧 𝐨𝐟 𝐬𝐭𝐚𝐭𝐞𝐬 𝐚𝐧𝐝 𝐜𝐨𝐯𝐚𝐫𝐢𝐚𝐧𝐜𝐞 : Dong Hyun Kim

Slide: https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch4-propagation-of-states-and-covariance

Youtube:https://youtu.be/38dVDay2XnA

𝐂𝐡 𝟓. 𝐓𝐡𝐞 𝐃𝐢𝐬𝐜𝐫𝐞𝐭𝐞-𝐓𝐢𝐦𝐞 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫: Jeong Eun Park

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch5-the-discretetime-kalman-filter

Youtube : https://www.youtube.com/watch?v=j9CRDGqmJ1I

𝐂𝐡 𝟔. 𝐀𝐥𝐭𝐞𝐫𝐧𝐚𝐭𝐞 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫 𝐅𝐨𝐫𝐦𝐮𝐥𝐚𝐭𝐢𝐨𝐧𝐬: Jin Hyuk Song

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch6-alternate-kalman-filter-formulations-jinhyuk-song

Youtube: https://www.youtube.com/watch?v=xtSLagWgG3g

𝐂𝐡 𝟕. 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫 𝐆𝐞𝐧𝐞𝐫𝐚𝐥𝐢𝐳𝐚𝐭𝐢𝐨𝐧𝐬 : Young Beom Kim

Slide :https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch7-kalman-filter-generalizations

Youtube :https://youtu.be/y1T4ogYlJb0

𝐂𝐡 𝟖. 𝐓𝐡𝐞 𝐂𝐨𝐧𝐭𝐢𝐧𝐮𝐨𝐮𝐬-𝐓𝐢𝐦𝐞 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫 : Hae Gu Lee

Slide :https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch8-the-continuoustime-kalman-filter

Youtube :https://youtu.be/HtS2SJcXcTE

𝐂𝐡 𝟗. 𝐎𝐩𝐭𝐢𝐦𝐚𝐥 𝐒𝐦𝐨𝐨𝐭𝐡𝐢𝐧𝐠: Hayden

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch9-optimal-smoothing-hayden

Youtube : https://youtu.be/Amm6pObx8No

𝐂𝐡 𝟏𝟎. 𝐀𝐝𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐓𝐨𝐩𝐢𝐜𝐬 𝐢𝐧 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫𝐢𝐧𝐠: Stella Seoyeon Yang

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch11-additional-topic-in-kalman-filter-stella-seoyeon-yang

Youtube: https://youtu.be/TI_fV4t6pOU

𝐂𝐡 𝟏𝟏. 𝐓𝐡𝐞 𝐇-𝐈𝐧𝐟𝐢𝐧𝐢𝐭𝐲 𝐅𝐢𝐥𝐭𝐞𝐫 : Young Beom Kim

Slide :https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch11-the-hinfinity-filter

Youtube :https://youtu.be/I_Kuj3coyFI

𝐂𝐡 𝟏𝟐. 𝐀𝐝𝐝𝐢𝐭𝐢𝐨𝐧𝐚𝐥 𝐓𝐨𝐩𝐢𝐜𝐬 𝐢𝐧 𝐇-𝐈𝐧𝐟𝐢𝐧𝐢𝐭𝐲 𝐅𝐢𝐥𝐭𝐞𝐫𝐢𝐧𝐠 : Hayden

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch12-additional-topics-in-hinfinity-filtering-hayden

Youtube :https://youtu.be/Y7xQXTdHKWw

𝐂𝐡 𝟏3. 𝐍𝐨𝐧𝐥𝐢𝐧𝐞𝐚𝐫 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫𝐢𝐧𝐠 : Sung Min Ahn

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch13-nonlinear-kalman-filtering-ahn-min-sung

Youtube :https://youtu.be/xucMwb7yAwY

𝐂𝐡 𝟏4. 𝐓𝐡𝐞 𝐔𝐧𝐬𝐜𝐞𝐧𝐭𝐞𝐝 𝐊𝐚𝐥𝐦𝐚𝐧 𝐅𝐢𝐥𝐭𝐞𝐫 : Soo Young Kim

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch14-the-unscented-kalman-filter-sooyoung-kimg

Youtube :https://youtu.be/niHpxUUbZ6c

𝐂𝐡 𝟏5. 𝐓𝐡𝐞 𝐏𝐚𝐫𝐭𝐢𝐜𝐥𝐞 𝐅𝐢𝐥𝐭𝐞𝐫 : Stella Seoyeon Yang

Slide : https://www.slideshare.net/roboticskrai/sensor-fusion-study-ch15-the-particle-filter-seoyeon-stella-yang

Youtube :https://youtu.be/23QPRsIlJHk

Additional Session :

𝐑𝐞𝐚𝐥𝐰𝐨𝐫𝐥𝐝 𝐒𝐞𝐧𝐬𝐨𝐫 𝐅𝐮𝐬𝐢𝐨𝐧 𝟏 : Soo Young Kim, Stella Seoyeon Yang

Slide1:https://www.slideshare.net/roboticskrai/sensor-fusion-study-real-world-1-lidar-radar-fusion-kim-soo-young-237932335

Slide2:https://www.slideshare.net/roboticskrai/sensor-fusion-study-real-world-2-gps-ins-fusion-stella-seoyeon-yang-237976578Youtube :https://youtu.be/23QPRsIlJHk

Youtube : https://youtu.be/74bdIvTIjvY

𝐑𝐞𝐚𝐥𝐰𝐨𝐫𝐥𝐝 𝐒𝐞𝐧𝐬𝐨𝐫 𝐅𝐮𝐬𝐢𝐨𝐧 𝟐 : Young Joo Kim, Kile

Youtube :https://youtu.be/t-4bAf3woSU

License

[Apache License 2.0][https://github.com/mercy-project/TLdroid/blob/develop/LICENSE]

Releases

No releases published

Packages

No packages published

© 2020 GitHub, Inc.

Terms

Privacy

Security

Status

Help

Contact GitHub

Pricing

API

Training

Blog

About





Posted by uniqueone
,

안녕하세요, TensorFlow KR 여러분!
StarGAN 리뷰 및 코드 실습 영상을 만들어 공유합니다. 한 장의 사진을 업로드해서 성별/나이/머리색 정보를 바꿔볼 수 있습니다. (구글 Colab에서 실행해 볼 수 있도록 만들어 놓았으므로, 관심이 있으신 분은 바로 코드부터 돌려보셔도 됩니다.)

코드 실습: https://github.com/ndb796/Deep-Learning-Paper-Review-and-Practice/blob/master/code_practices/StarGAN_Tutorial.ipynb

StarGAN은 Image-to-image Translation 분야에서 유명하고 오래된 (나온 지 3년 정도 된) 논문으로 TensorFlow KR에 계신 많은 분이 알고 계실 Sung Kim 교수님, 주재걸 교수님이 참여한 논문(최윤제님의)이기도 합니다.

현재는 StarGAN v2가 있으므로... StarGAN v2를 리뷰하려고 했으나 양이 너무 많아질 것 같아서 StarGAN v1부터 리뷰합니다. 워낙 말주변이 없어서... 간단하게 편집도 했는데도 1시간 20분짜리가 나왔네요... 필요하신 분께 도움이 되었으면 좋겠습니다.

https://www.youtube.com/watch?v=-r9M4Cj9o_8

Posted by uniqueone
,

Deep Reinforcement Learning Course (CS285) by Sergey Levine, Assistant Professor at University of California, Berkeley is now has fall 2020 lectures online:

#deeplearning #reinforcementlearning #RL #machinelearning


Posted by uniqueone
,

안녕하세요!

최근 모두의 말뭉치가 새로 공개된 겸하여 KoELECTRA-v3를 새로 만들었습니다:)
모두의 말뭉치를 포함한 약 34GB 데이터로 Pretraining
Mecab과 Wordpiece를 이용하여 Vocab 새로 제작

다행히도 여러 Task에서 눈에 띄는 성능 향상이 보이네요 (역시 데이터가 많은 게 최고네요ㅎ)
많은 분들이 사용해주셔서 항상 감사하고, 새로 공개한 모델도 한국어 NLP 연구에 많은 도움이 되었으면 좋겠습니다! 🤗
- Github: https://github.com/monologg/KoELECTRA

Posted by uniqueone
,

https://www.kdnuggets.com/2020/10/10-best-machine-learning-courses-2020.html



 


JMP Discovery Summit Online: Exploring Data, Inspiring Innovation. Oct 12-16. Register Now


Topics: Coronavirus | AI | Data Science | Deep Learning | Machine Learning | Python | R | Statistics


10 Best Machine Learning Courses in 2020

<= Previous post

Next post =>


 
 

Share24

Tags: CoursesDataCampDeep Learningfast.aiMachine LearningOnline EducationPythonStanford


If you are ready to take your career in machine learning to the next level, then these top 10 Machine Learning Courses covering both practical and theoretical work will help you excel.


SAS Viya



comments

By Ahmad Bin Shafiq, Machine Learning Student.

Photo by Photos Hobby on Unsplash.
 

Practical/Hands-on Courses with Less Theory

 
1) Practical Deep Learning for Coders FAST.AI

Price: Free
Taught by: One of the most famous and practical courses on the internet, taught by Jeremy Howard, Research Scientist at the University of San Francisco, the chair of WAMRI, and is Chief Scientist at the platform.ai. He was the President and Chief Scientist of the data science platform Kaggle, where he was the top-ranked participant in international machine learning competitions 2 years running.
Course Outcomes: This course is a hands-on introduction to deep learning, where you will dive straight into deep learning via making a state of the art classifier. You will learn a lot of practical aspects of deep learning without knowing the underlying theory.
 
2) Code-First Introduction to Natural Language Processing by Fast.ai

Price: Free
Taught by: Rachel Thomas is an American computer scientist and founding Director of the Center for Applied Data Ethics at the University of San Francisco. Together with Jeremy Howard, she is co-founder of fast.ai.
Course Outcomes: This course  is a hands-on introduction to NLP, where you will code a practical NLP application first as the name suggests, then slowly start digging inside the underlying theory in it.
Applications covered include topic modeling, classification (identifying whether the sentiment of a review is positive or negative), language modeling, and translation. The course teaches a blend of traditional NLP topics (including regex, SVD, naïve Bayes, tokenization) and recent neural network approaches (including RNNs, seq2seq, attention, and the transformer architecture), as well as addressing urgent ethical issues, such as bias and disinformation.
 
3) Python for Data Science and Machine Learning Bootcamp

Price: $129 (on sale $10-$20)
Taught by: Jose Marcial Portilla has a BS and MS in Mechanical Engineering from Santa Clara University and years of experience as a professional instructor and trainer for Data Science and programming. He has publications and patents in various fields such as microfluidics, materials science, and data science technologies.
Rating: 4.6*
Course Outcomes: This course is a very practical introduction to Machine Learning and data science. It does not assume any previous knowledge, starts from teaching basic Python to Numpy Pandas, then goes to teach Machine Learning via sci-kit learn in Python, then jumps to NLP and Tensorflow, and some big-data via spark.
This is definitely one of the best courses out there, as Jose is a really good instructor.
 
4) DeepLearning.AI TensorFlow Developer Professional Certificate

Price: $49/month
Taught by: Laurence Moroney is a Developer Advocate at Google working on Artificial Intelligence with TensorFlow. He is also the author of many books.
Rating: 4.7*
Course Outcomes: In this hands-on, four-course Professional Certificate program, you’ll learn the necessary tools to build scalable AI-powered applications with TensorFlow. Lawrence will start teaching you the basics of TensorFlow, slowly progressing towards the state of the art applications using Tensorflow.
 
5) Datacamp Data Science Path

Price: $25/month or $300/year
Taught by: Multiple industry professionals
Course Outcomes: With no prior coding experience, you will be taught coding from scratch, then moving to advanced libraries and frameworks. Each lesson is accompanied by some exercises or tasks. You will also have access to projects at data camp, which will improve your coding experience as well as your resume.
 

Theoretical Courses with Less Practical work

 
1) Machine Learning by Stanford University

Price: $80
Taught by: Andrew Ng is CEO/Founder Landing AI; Co-founder, Coursera; Adjunct Professor, Stanford University; formerly Chief Scientist, Baidu, and founding lead of Google Brain.
Rating:4.9
Course Outcomes: You will learn all the underlying theory behind famous machine learning algorithms, from Supervised Learning to Unsupervised Learning. You will also get a chance to code them from scratch in MATLAB/Octave.
 
2) Deep Learning Specialization

Price: $49/month
Instructor: Andrew Ng
Ratings: 4.8*
Course Outcomes: This 5 parts specialization will teach you the underlying theory behind of Deep Learning from Single Layer Network to Multi-Layer Dense Networks, from the basics of CNN to performing object detection with YOLO along with underlying theory, from basics of RNN to Sentiment analysis.
This course will also give you an introduction to the basics of Deep Learning frameworks such as Tensorflow or Keras.
 
3) CS231n by Andrej Karpathy

Price: Free
Taught by: Andrej Karpathy, the Sr. Director of AI at Tesla, leads the team responsible for all neural networks on the Autopilot. Previously, he was a Research Scientist at OpenAI working on Deep Learning in Computer Vision, Generative Modeling, and Reinforcement Learning. He received his Ph.D. from Stanford University.
Course Outcomes: This course is a deep dive into details of the deep learning architectures with a focus on learning end-to-end models for these tasks, particularly image classification. Students will learn to implement, train, and debug their own neural networks and gain a detailed understanding of cutting-edge research in computer vision. The focus is on teaching how to set up the problem of image recognition, the learning algorithms (e.g., backpropagation), practical engineering tricks for training, and fine-tuning the networks.
 
4) Stat 451: Introduction to Machine Learning

Price: Free
Taught by: Sebastian Raschka is an Assistant Professor of Statistics at the University of Wisconsin-Madison focusing on machine learning and deep learning research.
Course Outcomes: You will learn all the underlying theory of famous Machine Learning Algorithms from Neural Networks to supervised and Unsupervised Learning.
This course is originally taught at the University of Wisconsin-Madison by Dr. Sebastian.
 
5) MIT Introduction to Deep Learning | 6.S191

Price: Free
Taught by: Ava Soleimany is a Ph.D. student in the Harvard Biophysics program and at MIT, where she works with Sangeeta Bhatia at the Koch Institute for Integrative Cancer Research and am supported by the NSF Graduate Research Fellowship.
Alexander Amini is a Ph.D. student at MIT, in the Computer Science and Artificial Intelligence Laboratory (CSAIL), with Prof. Daniela Rus. He is an NSF Fellow and completed my Bachelor of Science and Master of Science in Electrical Engineering and Computer Science at MIT, with a minor in Mathematics.
Course Outcomes: 6.S191 is MIT’s official introductory course on deep learning methods with applications to computer vision, natural language processing, biology, and more! Students will gain foundational knowledge of deep learning algorithms.
Students will also get practical experience in building neural networks in TensorFlow.
 
Related:

The Online Courses You Must Take to be a Better Data Scientist

Online Certificates/Courses in AI, Data Science, Machine Learning from Top Universities

Machine Learning from Scratch: Free Online Textbook

<= Previous post

Next post =>




Top Stories Past 30 Days

Most Popular

Data Science Minimum: 10 Essential Skills You Need to Know to Start Doing Data Science

Introduction to Time Series Analysis in Python

Automating Every Aspect of Your Python Project

Free From MIT: Intro to Computer Science and Programming in Python

Machine Learning from Scratch: Free Online Textbook

If I had to start learning Data Science again, how would I do it?

Autograd: The Best Machine Learning Library Youre Not Using?

Most Shared

Data Science Minimum: 10 Essential Skills You Need to Know to Start Doing Data Science

Introduction to Time Series Analysis in Python

Modern Data Science Skills: 8 Categories, Core Skills, and Hot Skills

Machine Learning from Scratch: Free Online Textbook

Deep Learning’s Most Important Ideas

Free From MIT: Intro to Computer Science and Programming in Python

Statistics with Julia: The Free eBook

Latest News

Top tweets, Sep 30 – Oct 06: How to Explain Eac...

Free Introductory Machine Learning Course From Amazon

A step-by-step guide for creating an authentic data sci...

5 Challenges to Scaling Machine Learning Models

KDnuggets 20:n38, Oct 7: 10 Essential Skills You Need to Kn...

Effective Visualization Techniques for Data Discovery a...


Top Stories
Last WeekMost Popular

Data Science Minimum: 10 Essential Skills You Need to Know to Start Doing Data Science



The Best Free Data Science eBooks: 2020 Update

Geographical Plots with Python

Introduction to Time Series Analysis in Python

Comparing the Top Business Intelligence Tools: Power BI vs Tableau vs Qlik vs Domo

Machine Learning from Scratch: Free Online Textbook

5 Concepts Every Data Scientist Should Know


 
Most Shared

Data Science Minimum: 10 Essential Skills You Need to Know to Start Doing Data Science

The Best Free Data Science eBooks: 2020 Update

International alternatives to Kaggle for Data Science / Machine Learning competitions

Geographical Plots with Python

Machine Learning Model Deployment

How AI is Driving Innovation in Astronomy

The Online Courses You Must Take to be a Better Data Scientist

More Recent Stories

Effective Visualization Techniques for Data Discovery and Anal...

Here are the Most Popular Python IDEs/Editors

10 Best Machine Learning Courses in 2020

A Guide to Preparing OpenCV for Android

Top Stories, Sep 28 – Oct 4: Data Science Minimum: 10 Es...

New U. of Chicago Machine Learning for Cybersecurity Certifica...

Your Guide to Linear Regression Models

Key Machine Learning Technique: Nested Cross-Validation, Why a...

Getting Started in AI Research

Data Protection Techniques Needed to Guarantee Privacy

5 Concepts Every Data Scientist Should Know

Comparing the Top Business Intelligence Tools: Power BI vs Tab...

10 Days With “Deep Learning for Coders”

Data Science Minimum: 10 Essential Skills You Need to Know to Start Doing Data Science

Understanding Transformers, the Data Science Way

Top tweets, Sep 23-29: An Introduction to #AI – updat...

AI in Healthcare: A review of innovative startups

Machine Learning Model Deployment

The Best Free Data Science eBooks: 2020 Update

KDnuggets 20:n37, Sep 30: Introduction to Time Series Analy...

KDnuggets Home » News » 2020 » Oct » Tutorials, Overviews » 10 Best Machine Learning Courses in 2020 ( 20:n38 )



© 2020 KDnuggets. | About KDnuggets  | Contact  | Privacy policy  | Terms of Service
 

41SHARES


Posted by uniqueone
,

Made With ML Topics

A collection of the best ML tutorials, toolkits and research organized by topic: https://madewithml.com/topics/

H / T : Goku Mohandas

#DeepLearning #MachineLearning #Tutorials

Posted by uniqueone
,

https://www.facebook.com/groups/DeepAI/permalink/2577160382575423/?sfnsn=mo

Facebook 그룹

AI Deep에 멤버 22,364명이 있습니다. Promise me you’ll always remember: You’re braver than you believe, and stronger than you seem, and smarter than you think. Follow us on :...

www.facebook.com

Having an unbalanced dataset is not an uncommon problem in real world. Interested in knowing how to handle Imbalanced Classification problem? Checkout this brief introduction from team at Robofied.

#AI #ML #artificialintelligence #machinelearning #DL #deeplearning #neuralnetworks #dataanalytics #datascience #data #bigdata #technology #automation #datascientist #programming #code #coding #developer #tech #love #geek #ecommerce #design #startup #contentmarketing #branding #business #robofied

Posted by uniqueone
,

Multi-View Brain HyperConnectome AutoEncoder For Brain State Classification

Banka et al.: https://arxiv.org/abs/2009.11553

#DeepLearning #HyperConnectome #AutoEncoder

Posted by uniqueone
,

State of the art in 3D dense face alignment!
For project and code/API/expert requests: https://www.catalyzex.com/paper/arxiv:2009.09960

Posted by uniqueone
,

numpy.r_ : Translates slice objects to concatenation along the first axis. 첫번째 axis (위에서 아래 방향)으로 연결되도록 변환한다.

numpy.c_ : Translates slice objects to concatenation along the second axis. 두번째 axis (왼쪽에서 오른 방향)으로 연결되도록 변환한다.

 

>>> import numpy as np

>>> a = np.array([123]) # 1차원

>>> b = np.array([456]) # 1차원

>>> c = np.array([[123]]) # 2차원

>>> d = np.array([[456]]) # 2차원

>>> e = np.array([[[123]]]) # 3차원

 

>>> a.shape, b.shape, c.shape, d.shape, e.shape

((3,), (3,), (1, 3), (1, 3), (1, 1, 3))

 

>>> np.r_[a, b]

array([1, 2, 3, 4, 5, 6])

a, b는 1차원 어레이인데, 위아래로 합치면 [1, 2, 3, 4, 5, 6]^T 이 되는데 (즉 세로 형태의 1차원 array가 되는데),  파이썬에서는 1차원 어레이는 무조건 가로로 나열되므로 결과는 array([1, 2, 3, 4, 5, 6])가 된다.

 

>>> np.r_[c, d]

array([[1, 2, 3], [4, 5, 6]])

c, d는 2차원 어레이인데, 위아래로 합치면 array([[1, 2, 3], [4, 5, 6]])가 된다. 

 

>>> np.c_[a, b]

array([[1, 4], [2, 5], [3, 6]])

a, b는 1차원 어레이인데, 수평으로 합치면 array([[1, 4], [2, 5], [3, 6]])가 된다.

 

>>> np.c_[c, d]

array([[1, 2, 3, 4, 5, 6]])

c, d는 2차원 어레이인데, 수평으로 합치면 array([[1, 2, 3, 4, 5, 6]])가 된다. 

 

파이썬python_블로그에설명.pptx
0.04MB

 

numpy.org/doc/stable/reference/generated/numpy.r_.html?highlight=r_#numpy.r_

rfriend.tistory.com/352

 

 

 

 

 

 

 

 

 
 
 
 
Posted by uniqueone
,

인터넷에서 30종 인기 블로그나 온라인 사이트를 정리해 공유하였습니다. 참고가 되었으면 좋겠습니다~

 Toggle Navigation

/* Template Name: singlenew(2019.11.12) */    



빅데이터 분석 : 책보다 편한 30종 스터디 사이트

최종 업데이트:2020-5-18


데이터 과학에 관한 작업을 시작했을 뿐이지만 데이터 분석과 빅데이터 분석 전문가(ADP) 라이선스 준비를 위해 전문 서적 외에도 관련 사이트나 블로그로부터 최신 데이터 정보와 지식을 얻어야 합니다. 인터넷에서 30종 인기 블로그나 온라인 사이트를 정리해 공유하였습니다. 참고가 되었으면 좋겠습니다.
실제로 빅데이터 분석 을 할 때 좋은 정보와 데이터가 없는 경우가 많습니다. 오픈 공공데이터 포럼을 정리하고 소개하였습니다. 2 가지 함께 보시면 더 좋을 것 같습니다.

빅데이터 분석 : 온라인 사이트

  1.Dataquest:
https://www.dataquest.io/home


이 빅데이터 분석 사이트는 데이터 사이언스트,데이터 애널리스트, 또는 데이터 엔지니어에 대한 관심을 가지는 분들은 대화형 코딩 과제부터 공부하고 기준 문제 및 데이터 사이언스 프로젝트로 성장할 수 있습니다.

2.Udemy:
https://www.udemy.com


세계 최대급의 온라인 빅데이터 분석 학습 플랫폼으로서 알려졌으며 이제 8000이상의 온라인 코스가 있습니다. 머신러닝(machine learning)부터 데이터 시각화, 그리고 Python 프로그래밍까지를 포함하고 있습니다. 

3.Coursera:
https://www.coursera.org/browse/data-science?languages=en


Coursera는 전 세계 대학 및 기관과 협력하여 온하인 코스를 제공하고 있습니다.  등록자는 온라인 비디오 빅데이터 분석 강의에 엑셀을 할 수 있고, 숙제가 자동 평가되고, 커뮤니티 포럼에도 들어갈 수 있습니다. 코스 종료 후에 전자 수료증명서가 발급됩니다.

4.DataCamp:
 httpswww.datacamp.com


 스킬을 향상시키고 싶은 초보자나 숙련된 전문가에게  되게 훌륭한 학습 자원이라고 할 수 있습니다. DataCamp는 데이터 사이언스, 통계학 및 머신러닝에 관련된 인터랙티브한 R 및 Python 코스를 제공하고 있습니다.

5.Udacity:
https://www.udacity.com/course/data-analyst-nanodegree–nd002


Udacity는 Python、SQL 및 통계 데이터를 통해 인사이트를 통찰하고 중요한 발견을 전하고 데이터 구동형(driven) 솔루션을 구축하기 위해서 힘을 애쓰고 있다는 슬로건이 있습니다. 온라인 비디오 뿐만 아니라, 독자적인 스터디 관리 시스템, 빌트인 (built-in) 프로그래밍 인터페이스 및 포럼도 가지고 있습니다.

6.EdX:
https://www.edx.org/learn/data-science


 EdX에서는 데이터 분석 스킬을 기르고, 지식을 높이기 위한 무료 데이터 사이언스 코스를 제공하고 있습니다. 전 세계의 답대학 및 EdX의 글로벌 파트너의 코스를 통해 데이터 사이언스, 통계, 비즈니스 인텔리전스(BI)에 관련 자료를 함계 공부합시다.

7.CognitiveClass:
https://cognitiveclass.ai


이 사이트에서 데이터 사이언스, 인공지능, 빅데이터, 블록체인 스킬을 무료로 수강할 수 있습니다. 안드로이드 전용의 앱이 있기 때문에 사용하기 편리한다는 장점이 있습니다.

8.Springboard:
https://www.springboard.com/workshops/data-science-career-track?afmc=2h


여기서 데이터 사이언스 기초 지식을 섭득하고, 실제에 따라 high-level의 머신러닝, 자연 언어 처리, 딥러닝 등의 중점 분야를 선택하고 공부할 수 있습니다.  실제 프로젝트 연습 뿐만 아니라 맟춤형 캐리어 코칭 및 구인 사이트도 제공하고 있습니다. 

 9.Data School(YouTube):
https://www.youtube.com/user/dataschool/featured


Data School는 워싱턴DC에서 데이터 사이언스 교육을 하고 있는 Kevin Markham씨의 채널입니다. 이 채널에서 오픈 소스 툴에 관련 사용방법, Python 및 R 등 다양한 데이터 사이언스에 기반으로 주제에 대해서 소개하는 것입니다.  무료 보기 가능하고, 현재 8만명 이상의 팔로워가 가지고 있습니다.

10.Siraj Raval(YouTube) :
https://www.youtube.com/channel/UCWN3xxRkmTPmbKwht9FuE5A/featured


Siraj Raval씨의 채널은 개발자에게 인공지능 구축을 교육하기 위한 비디오를 제고하고 있는 것입니다. 게임,음악, 채팅봇, 아트 등을 AI로 만드는 과정을 시청자들에게 알려주었습니다. 이제 팔로워는 50만명을 넘고 있습니다.

11.IntelliPaat:
https://intellipaat.com/tutorials/


IntelliPaat의 튜토리얼은 빅데이터 , 비즈니스 인텔리전스, 데이터 분석, 데이터베이스등 테마를 커버하고 있습니다. 업계 전문가에게 수강을 수료한다면 인정(증면서)를 받을 수 있습니다. 

12.Lynda:
https://www.lynda.com/search?q=data+science


Lynda에서는 빅데이터, 비즈니스 인테리전스, 데이터 분석, 데이터베이스,프로그래밍 언어 등이 포함됩니다. 자신의 수준에 따라 입문, 중급, 상급 코스를 선택할 수 있습니다. 

13.Kaggle:
https://www.kaggle.com/learn/overview


Kaggle는 초보자에게 최우선 선택이라고 할 수 있습니다. Python,머신러닝, 데이터 시각화, Pandas, R, SQL 또는 딥러닝에 관련 코스를 선택할 수 있습니다.


데이터 사이언스 블로그

1. EdwinChen 개안 블로그
http://blog.echen.me


EdwinChen 씨는 매사추세츠공과대학(MIT)에서 수학과 언어학을 전공했으며, MRS음성인식(speech recognition),Clarium의 양적 거래, Twitter 광고, Dropbox 분석, Google 데이터 사이언스 등의 일을 하고 있었습니다. AI 인공지능과 데이터에 관한 포스트를  볼 수 있습니다.

 2.DataScience.com:
https://www.datascience.com/blog


이 블로그는 데이터 사이언티스트, 비즈니스 애널리스트 및 IT팀을 대상으로 운영하고 있습니다. 그리고 이에 대한 자신의 생각을 전하고 싶다면 여기에 독자들의 투고를 환영합니다. 

3.데이터 사이언스/101
http://101.datascience.community


컴퓨터 과학 및 통계 과학 박사 학위를 취득한 Ryan Swanstrom는 2012년부터 자신의  블로그에서 데이터 과학 자원, 뉴스 및 연구에 대한 뉴스를 공개하고 있습니다.

 4.The Shapeof Data:
https:shapeofdata.wordpress.com


JesseJohnson은 저자원 토폴로지를 연구하던 전직 수학 교수로 현재 구글 소프트워어 엔지니를 하고 있습니다. 블로그에서 현대 데이터 분석의 기본적인 사고방식에 대해 탐색하고 성명하고 있습니다. 

 5.PlanetBig:
http://planetbigdata.com


PlanetBig Matoop는 빅데이터 및 Hadoop등에 관련 블로그를 정리하는 플랫폼입니다. 전세계의 블로그를 모았습니다.

6. 빅데이터 블로그:
http://blog.bigdataweek.com


Big Data Week 회의에서의 인터뷰와 의견을 공개하고 있습니다. 회의에 참석할 수 없는 경우에는 이 블로그에서 최신 정보를 입수할 수 있습니다.

7.ChrisAblon
https://chrisalbon.com/


ChrisAblon은 통계,인공지능 및 소포트웨어 분약에서 10년이상 경험을 가진 데이터 과학자 및 정치학자입니다. 정치, 사회, 인도 활동에 익숙하며 이에 대한 많은 자료를 공유하고 있습니다.

 8.NYC AcienceAcademyAcademy
https://nycdatascience.com/blog/


교육기관인 NYC Data Science Academy의 블로그에서 R, Web 크롤, 머신러닝에 관한 포스트를 공개하고 있으며 온라인 코스도 있습니다. 데이터 사이언스 분야에서 축적하는 경험에 의해서, 기업에 스킬과 전략 트레이닝을 제공하고 있습니다.

9.DBMS2:
http://www.dbms2.com/


DBMS2는 데이터베이스와 데이터 분석 스킬에 관심이 있는 분들에게 업계 전문가의 리뷰를 제공하는 플랫폼입니다. 필자인 Curt Monash씨는 30년 동안 업계에 주목하고 왔습니다.

10.Data36:
https://data36.com/


Tomi Mester씨의 블로그에서는, 온라인 데이터 아날리스트 베스트 프랙티스에 대해 자세하게 해설하고 있습니다. 데이터 분석, AB테스트, 조사연구 및 데이터 사이언스에 관련된 포스트, 코스 및 비디오를 공유하고 있습니다.

11.데이터베이스 관리시스템 운영:
http://www.odbms.org/


이 사이트는 빅데이터, 최신 데이터 관리 테크놀로지 그리고 데이터 사이언스에 관련 자료가 포함된 포털입니다. 필자는 프랑크푸르트 대학의 데이터베이스 및 정보 시스템 교수입니다.

12.Yanir Seroussi 블로그 :
https://yanirseroussi.com/


Yanir Seoussi씨는 프로그래밍, 컴퓨터 사이언스, 머신러닝 및 통계에서 풍부한 경험을 가진 데이터 사이언티스트 및 소프트웨어 엔지니어입니다.그의 블로그에 고립된 데이터의 문제부터 생산 시스템의 구축까지의 아이디어가 포함되어 있습니다.

13.SimplyStatistics:
https://simplystatistics.org/


이 블로그는 Jeff Leek, Roger Peng, Rafa Irizarry 3명의 생물 통계학 교수에 의해서 쓰여졌습니다. 과학적 논의에 기여하는 흥미로운 포스트에 대해 코멘트하고, 그리고 신인 통계학자와 의견을 공유합니다.

 14.What is big data
https://whatsthebigdata.com/


Gil Press는 라이팅, 리서치, 마케팅 서비스를 제공하는 “gPress”라는 컨설팅 사업을 경영하고 있습니다.  블로그에서는 빅데이터에 관한 정보기술, 비즈니스 업계, 정부기관, 그리고 우리 생활에 미치는 영향에 대한 것을 알아보고 있습니다.

15.Towards Data Science:
https://towardsdatascience.com/


이 블로그는 수천명의 사람들이 의견을 교환하고 데이터 사이언스에 대한 이해를 증진시키는 플랫폼입니다.각 전문적인  데이터 사이언티스트로부터 다양한 내용을 읽을 수 있습니다.

 16.AlexisPerrier:
https://alexisperrier.com/


Alexis Perrier씨는 기업의 머신러닝을 지원한 적이 있는 데이터 사이언스 강사입니다. 자신의 블로그에서 선형회귀 등에 대한 상세한 연구를 공유했습니다.
러닝을 지원한 적이 있는 데이터 사이언스 강사입니다. 자신의 블로그에서 선형회귀 등에 대한 상세한 연구를 공유했습니다.

17.Algobeans:
https://algobeans.com/


Algobeans는 케임브리지대학의 Annalyn 및 스탠퍼드대학의 Kenneth에 의해 설립되었습니다. 데이터 사이언스의 애호가로서 누구나 쉽게 데이터 사이언스를 공부할 수 있도록 사이트를 작성했습니다.
페이스북에서 FineReport Reporting Sofeware채널을 구독하고 더 많은 데이터 시각화 정보를 받으세요!

체험판 무료로 알아보기

더 많은 보기…

1분 안에 인사으트를 보는 대시보드 디자인 도구 및 제작사례 10가지
2020년 BI Tool의 주요 기능 TOP10
보고서 유형 및 샘플 10가지 알아보기!

FacebookTwitterEmail나누기

2020년 국내외 데이터 시각화 툴 4종 사용 후기



2020년 은행 대시보드 인기 샘플 TOP 3



위로

제품문의
상담/문의: 031-450-3011
고객 지원:korea@fanruan.com

               
Copyright©2018 FanRuan Software Co., Ltd.
苏ICP备14031611号-1


Posted by uniqueone
,

PR(Precision Recall) curve VS ROC curve에 대한 근본적인 이해를 돕는 글들이 있다. 

                       | predicted negative | predicted positive |
|-----------------|---------------------|---------------------|
| actual negative | True Negative (TN) | False Positive (FP) |
| actual positive  | False Negative (FN) | True Positive (TP) |

 

[1]과 [2]에서 설명하는 내용은 비슷하다. ROC 커브는 Recall(=Sensitivity)과  False Positive Rate(= 1-Specificity)인데, 이 둘은 모두 actual class가 1인 것중에서 1로 분류한 비율과 0인 것 중에서 1로 분류한 것의 비율을 보는 것인데, PR 커브는 Precision과 Recall(=Sensitivity)을 사용하는데, Recall은 같은데 Precision은 1로 분류된 것 중에서 실제 1의 비율을 보는 것이다. Precision은 실제클래스가 1인 샘플들의 base probability 또는 prior probability가 달라짐에 따라서 Precision이 달라진다. 반면, Recall(=Sensitivity)과 FPR은 base probability 또는 prior probability가 항상 같다. [3]에 PR, ROC를 잘 설명한 그림이 있어 복사해왔다. 

 

src: https://towardsdatascience.com/on-roc-and-precision-recall-curves-c23e9b63820c

 

위 그림을 예로들어 설명하면 (분홍색 선과 청록색 선은 threshold를 나타냄), 

Recall(=Sensitivity) : 전체 노란 샘플개수 중 분홍색 선 위쪽에 존재하는 노란 샘플개수 비율.

False Positive Rate(= 1-Specificity) : 전체 파란 샘플개수 중 분홍색 선 위쪽에 존재하는 파란 샘플개수 비율.

Precision : 분홍색 선 위쪽에 존재하는 노란색 샘플과 파란색 샘플의 합 중에서 분홍색 선 위쪽에 존재하는 노란 샘플개수 비율.

 

Recall(=Sensitivity)과 FPR은 threshold가 청록cyan색에서 분홍색 라인으로 올라가도 분자는 바뀌어도 분모(base probability 또는 prior probability)는 그대로이다. 반면, Precision은 threshold가 바뀌면 분모(base probability 또는 prior probability)가 바뀌고 분자도 바뀐다. 위와 같이 imbalanced data인 경우 threshold가 위로 올라가면서 Precision의 분모가 급격히 바뀌게 된다. 위 그림에서 클래스 1인 노란색 샘플이 파란색 샘플만큼 많은 경우를 상상해보자. 즉 파란색 샘플들은 그대로인데, 노란색 샘플들이 prediction score가 대략 [-0.3, 1] 범위에 노란 샘플들이 많이 존재하는 경우다. 그러면 threshold가 올라가면서 Precision의 분모가 급격히 변하지 않고 smooth하게 변할 것이다. 윗부분이 노란색 샘플들이 많으므로. 그러면 threshold가 위로 갈수록 분모와 분자는 거의 비슷한 값을 가질 것이다. threshold가 변함에 따라 Precision이 급격하게 변하지 않는 것이다. 따라서 Precision은 1클래스가 rare한 경우에 민감하여 분류 performance를 잘 반영해준다. 

 

 

[1] stats.stackexchange.com/questions/7207/roc-vs-precision-and-recall-curves

 

The key difference is that ROC curves will be the same no matter what the baseline probability is, but PR curves may be more useful in practice for needle-in-haystack type problems or problems where the "positive" class is more interesting than the negative class.

To show this, first let's start with a very nice way to define precision, recall and specificity. Assume you have a "positive" class called 1 and a "negative" class called 0. Y^Y^ is your estimate of the true class label YY. Then:

 

Precision            =P(Y=1|Y^=1)

Recall=Sensitivity =P(Y^=1|Y=1)

Specificity           =P(Y^=0|Y=0)

 

The key thing to note is that sensitivity/recall and specificity, which make up the ROC curve, are probabilities conditioned on the true class label. Therefore, they will be the same regardless of what P(Y=1)P(Y=1) is. Precision is a probability conditioned on your estimate of the class label and will thus vary if you try your classifier in different populations with different baseline P(Y=1). However, it may be more useful in practice if you only care about one population with known background probability and the "positive" class is much more interesting than the "negative" class. (IIRC precision is popular in the document retrieval field, where this is the case.) This is because it directly answers the question, "What is the probability that this is a real hit given my classifier says it is?".

Interestingly, by Bayes' theorem you can work out cases where specificity can be very high and precision very low simultaneously. All you have to do is assume P(Y=1) is very close to zero. In practice I've developed several classifiers with this performance characteristic when searching for needles in DNA sequence haystacks.

IMHO when writing a paper you should provide whichever curve answers the question you want answered (or whichever one is more favorable to your method, if you're cynical). If your question is: "How meaningful is a positive result from my classifier given the baseline probabilities of my problem?", use a PR curve. If your question is, "How well can this classifier be expected to perform in general, at a variety of different baseline probabilities?", go with a ROC curve.

 

[2] www.quora.com/What-is-the-difference-between-a-ROC-curve-and-a-precision-recall-curve-When-should-I-use-each

 

There is a very important difference between what a ROC curve represents vs that of a PRECISION vs RECALL curve.

Remember, a ROC curve represents a relation between sensitivity (RECALL) and False Positive Rate (NOT PRECISION). Sensitivity is the other name for recall but the False Positive Rate is not PRECISION.

Recall/Sensitivity is the measure of the probability that your estimate is 1 given all the samples whose true class label is 1. It is a measure of how many of the positive samples have been identified as being positive.

Specificity is the measure of the probability that your estimate is 0 given all the samples whose true class label is 0. It is a measure of how many of the negative samples have been identified as being negative.

PRECISION on the other hand is different. It is a measure of the probability that a sample is a true positive class given that your classifier said it is positive. It is a measure of how many of the samples predicted by the classifier as positive is indeed positive. Note here that this changes when the base probability or prior probability of the positive class changes. Which means PRECISION depends on how rare is the positive class. In other words, it is used when positive class is more interesting than the negative class.

So, if your problem involves kind of searching a needle in the haystack when for ex: the positive class samples are very rare compared to the negative classes, use a precision recall curve. Othwerwise use a ROC curve because a ROC curve remains the same regardless of the baseline prior probability of your positive class (the important rare class).

 

 

[3] towardsdatascience.com/on-roc-and-precision-recall-curves-c23e9b63820c

 
 
 
 
 
 
Posted by uniqueone
,

풀스택 딥러닝

[https://course.fullstackdeeplearning.com/](https://course.fullstackdeeplearning.com/?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

예전에 어떤분이 텐플코에 공유해주셔서 꼭 한번 봐야겠다 생각해서 3주정도 천천히 보면서 내용을 정리해서 포스팅해봤습니다! 유명한 게스트 강사들의 강연도 합치면 10시간치 영상은 더 있는 것 같지만 그건 내용 정리에 대한 부담없이 편하게 보려고 합니다.

머신러닝을 배우다보면 모델 학습과 아키텍처에 치중되게 되는데, 상대적으로 부족한 프로덕션에서 필요한 머신러닝 팀과 프로젝트 관리, 데이터 파이프라인 관리, 모델 배포 등에 대해서 잘 정리된 강의입니다.

제가 잘 몰라서 들으면서 잘 몰라가지고 ?? 쳐놓으며 써놓은 부분도 있는데 ㅠ 그런 부분들 댓글로 알려주시면 감사하겠습니다.

1. Setting up Machine Learning Projects [https://medium.com/@heegyukim/full-stack-deep-learning-setting-up-machine-learning-projects-32681ccac007](https://medium.com/@heegyukim/full-stack-deep-learning-setting-up-machine-learning-projects-32681ccac007?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

2. Infrastructure and Tooling

[https://medium.com/@heegyukim/full-stack-deep-learning-infrastructure-and-tooling-e0cd33af7a9e](https://medium.com/@heegyukim/full-stack-deep-learning-infrastructure-and-tooling-e0cd33af7a9e?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

3. Data Management

[https://medium.com/@heegyukim/full-stack-deep-learning-data-management-7ee8ca43aac7](https://medium.com/@heegyukim/full-stack-deep-learning-data-management-7ee8ca43aac7?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

4. Machine Learning Teams

[https://medium.com/@heegyukim/full-stack-deep-learning-machine-learning-teams-8aaa2e22b5ca](https://medium.com/@heegyukim/full-stack-deep-learning-machine-learning-teams-8aaa2e22b5ca?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

5. Training and Debugging

[https://medium.com/@heegyukim/full-stack-deep-learning-training-and-debugging-1c0f1d587e66](https://medium.com/@heegyukim/full-stack-deep-learning-training-and-debugging-1c0f1d587e66?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

6. Testing and Deployment

[https://medium.com/@heegyukim/full-stack-deep-learning-testing-and-deployment-d98568424a8a](https://medium.com/@heegyukim/full-stack-deep-learning-testing-and-deployment-d98568424a8a?fbclid=IwAR0tUATP_PY5_xNXnQv4s0D3xaozZvuHxqrjCY2r37Em6bj4i8ATqf7C8uQ)

Posted by uniqueone
,

Posted by uniqueone
,

https://youtu.be/FeBhlDLAddc
Converting video (mp4) file to audio (mp3) file using python.
converting video to audio without using any third-party software.
Python tutorial.
#python #pythoncode #pythonprogramming #pythonlearning

Posted by uniqueone
,

https://www.facebook.com/groups/PyTorchKR/permalink/1826343574171982/

Facebook 그룹

PyTorch KR에 멤버 10,039명이 있습니다. PyTorch를 이용한 자유로운 머신러닝 이야기의 장, PyTorch 한국 사용자 그룹 PyTorch KR입니다. PyTorch KR slack 가입 링크: https://pytorch-kr-invitation.herokuapp.com PyTorch KR sla

www.facebook.com


안녕하세요! super resolution과 관련한 정말 기초적인 질문이 있어 왔습니다!ㅜ.ㅜ

이미지 분류는 resnet과 같은 CNN 기반의 모델, detection은 yolo 등과 같이 각 task의 대표적인(상징적인?) 모델이 있습니다.(sota가 아니더라도..)

Super resolution에서는 어떤 모델이 대표적인, 혹은 많이 사용되는지 궁금합니다!

가장 대표적인 모델은 SRGAN이 아닐까 싶습니다
https://arxiv.org/abs/1609.04802

https://hoya012.github.io/blog/SIngle-Image-Super-Resolution-Overview/

제 블로그에 Super Resolution 관련하여 정리해둔게 있어서 참고하시면 좋을 것 같습니다.



Posted by uniqueone
,

Bayesian Data Analysis course
Vehtari et al., Aalto (CS-E5710) : https://avehtari.github.io/BDA_course_Aalto/
#Bayesian #DataAnalysis #MarkovChainMonteCarlo

Prerequisites

Course contents following BDA3

How to study

Slides and chapter notes

Videos

R and Python

Demos

Self study exercises

Stan

Extra reading

Acknowledgements

Bayesian Data Analysis course

Bayesian Data Analysis course

Page updated: 2020-08-24

This is the web page for the Bayesian Data Analysis course at Aalto (CS-E5710) by Aki Vehtari.
Aalto students should check also MyCourses announcements. In Autumn 2020 the course will be arranged completely online. This web page will be much updated during the August.
All the course material is available in a git repo (and these pages are for easier navigation). All the material can be used in other courses. Text (except the BDA3 book) and videos licensed under CC-BY-NC 4.0. Code licensed under BSD-3.



The electronic version of the course book Bayesian Data Analysis, 3rd ed, by by Andrew Gelman, John Carlin, Hal Stern, David Dunson, Aki Vehtari, and Donald Rubin is available for non-commercial purposes. Hard copies are available from the publisher and many book stores. See also home page for the bookerrata for the book, and chapter notes.

Prerequisites

Basic terms of probability theory

probability, probability density, distribution

sum, product rule, and Bayes' rule

expectation, mean, variance, median

in Finnish, see e.g. Stokastiikka ja tilastollinen ajattelu

in English, see e.g. Wikipedia and Introduction to probability and statistics

Some algebra and calculus

Basic visualisation techniques (R or Python)

histogram, density plot, scatter plot

see e.g. BDA R demos

see e.g. BDA Python demos

This course has been designed so that there is strong emphasis in computational aspects of Bayesian data analysis and using the latest computational tools.
If you find BDA3 too difficult to start with, I recommend

For regression models, their connection to statistical testing and causal analysis see Gelman, Hill and Vehtari, "Regression and Other Stories".

Richard McElreath's Statistical Rethinking, 2nd ed book is easier than BDA3 and the 2nd ed is excellent. Statistical Rethinking doesn't go as deep in some details, math, algorithms and programming as BDA course. Richard's lecture videos of Statistical Rethinking: A Bayesian Course Using R and Stan are highly recommended even if you are following BDA3.

For background prerequisites some students have found chapters 2, 4 and 5 in Kruschke, "Doing Bayesian Data Analysis" useful.

Course contents following BDA3

Bayesian Data Analysis, 3rd ed, by by Andrew Gelman, John Carlin, Hal Stern, David Dunson, Aki Vehtari, and Donald Rubin. Home page for the bookErrata for the bookElectronic edition for non-commercial purposes only.

Background (Ch 1, Lecture 1)

Single-parameter models (Ch 2, Lecture 2)

Multiparameter models (Ch 3, Lecture 3)

Computational methods (Ch 10 , Lecture 4)

Markov chain Monte Carlo (Chs 11-12, Lectures 5-6)

Extra material for Stan and probabilistic programming (see below, Lecture 6)

Hierarchical models (Ch 5, Lecture 7)

Model checking (Ch 6, Lectures 8-9)

Visualization in Bayesian workflow

Evaluating and comparing models (Ch 7)

Practical Bayesian model evaluation using leave-one-out cross-validation and WAIC (Journal link)

Videos and case studies

Cross-validation FAQ

Decision analysis (Ch 9, Lecture 10)

Large sample properties and Laplace approximation (Ch 4, Lecture 11-12)

In addition you learn workflow for Bayesian data analysis

How to study

Recommended way to go through the material is

Read the reading instructions for a chapter in chapter notes.

Read the chapter in BDA3 and check that you find the terms listed in the reading instructions.

Watch the corresponding lecture video to get explanations for most important parts.

Read corresponding additional information in the chapter notes.

Run the corresponding demos in R demos or Python demos.

Read the exercise instructions and make the corresponding assignments. Demo codes in R demos and Python demos have a lot of useful examples for handling data and plotting figures. If you have problems, visit TA sessions or ask in course slack channel.

If you want to learn more, make also self study exercises listed below

Slides and chapter notes

Slides

including code for reproducing some of the figures

Chapter notes

including reading instructions highlighting most important parts and terms

Videos

The following video motivates why computational probabilistic methods and probabilistic programming are important part of modern Bayesian data analysis.

Computational probabilistic modeling in 15mins

Short video clips on selected introductory topics are available in a Panopto folder and listed below.

1.1 Introduction to uncertainty and modelling

1.2 Introduction to the course contents

2.1 Observation model, likelihood, posterior and binomial model

2.2 Predictive distribution and benefit of integration

2.3 Priors and prior information

2019 fall lecture videos are in a Panopto folder and listed below.

Lecture 2.1 and Lecture 2.2 on basics of Bayesian inference, observation model, likelihood, posterior and binomial model, predictive distribution and benefit of integration, priors and prior information, and one parameter normal model. BDA3 Ch 1+2.

Lecture 3 on multiparameter models, joint, marginal and conditional distribution, normal model, bioassay example, grid sampling and grid evaluation. BDA3 Ch 3.

Lecture 4.1 on numerical issues, Monte Carlo, how many simulation draws are needed, how many digits to report, and Lecture 4.2 on direct simulation, curse of dimensionality, rejection sampling, and importance sampling. BDA3 Ch 10.

Lecture 5.1 on Markov chain Monte Carlo, Gibbs sampling, Metropolis algorithm, and Lecture 5.2 on warm-up, convergence diagnostics, R-hat, and effective sample size. BDA3 Ch 11.

Lecture 6.1 on HMC, NUTS, dynamic HMC and HMC specific convergence diagnostics, and Lecture 6.2 on probabilistic programming and Stan. BDA3 Ch 12 + extra material.

Lecture 7.1 on hierarchical models, and Lecture 7.2 on exchangeability. BDA3 Ch 5.

Project work info

Lecture 8.1 on model checking, and Lecture 8.2 on cross-validation part 1. BDA3 Ch 6-7 + extra material.

Lecture 9.1 PSIS-LOO and K-fold-CV, Lecture 9.2 model comparison and selection, and Lecture 9.3 extra lecture on variable selection with projection predictive variable selection. Extra material.

Lecture 10.1 on decision analysis. BDA3 Ch 9.

Project presentation info

Lecture 11.1 on normal approximation (Laplace approximation) and Lecture 11.2 on large sample theory and counter examples. BDA3 Ch 4.

Lecture 12.1 on frequency evaluation, hypothesis testing and variable selection and Lecture 12.2 overview of modeling data collection, BDA3 Ch 8, linear models, BDA Ch 14-18, lasso, horseshoe and Gaussian processes, BDA3 Ch 21.

R and Python

We strongly recommend using R in the course as there are more packages for Stan and statistical analysis in R. If you are already fluent in Python, but not in R, then using Python may be easier, but it can still be more useful to learn also R. Unless you are already experienced and have figured out your preferred way to work with R, we recommend installing RStudio Desktop. See FAQ for frequently asked questions about R problems in this course. The demo codes provide useful starting points for all the assignments.

For learning R programming basics we recommend

Garrett Grolemund, Hands-On Programming with R

For learning basic and advanced plotting using R we recommend

Kieran Healy, Data Visualization - A practical introduction

Antony Unwin, Graphical Data Analysis with R

Demos

R demos

Python demos

Self study exercises

Great self study BDA3 exercises for this course are listed below. Most of these have also model solutions available.

1.1-1.4, 1.6-1.8 (model solutions for 1.1-1.6)

2.1-2.5, 2.8, 2.9, 2.14, 2.17, 2.22 (model solutions for 2.1-2.5, 2.7-2.13, 2.16, 2.17, 2.20, and 2.14 is in slides)

3.2, 3.3, 3.9 (model solutions for 3.1-3.3, 3.5, 3.9, 3.10)

4.2, 4.4, 4.6 (model solutions for 3.2-3.4, 3.6, 3.7, 3.9, 3.10)

5.1, 5.2 (model solutions for 5.3-5.5, 5.7-5.12)

6.1 (model solutions for 6.1, 6.5-6.7)

9.1

10.1, 10.2 (model solution for 10.4)

11.1 (model solution for 11.1)

Stan

Stan home page

Introductory article in Journal of Statistical Software

Documentation

RStan installation

PyStan installation

Basics of Bayesian inference and Stan, Jonah Gabry & Lauren Kennedy Part 1 and Part 2

Extra reading

Dicing with the unknown

Origin of word Bayesian

Model selection

Cross-validation FAQ

Acknowledgements

The course material has been greatly improved by the previous and current course assistants (in alphabetical order): Michael Riis Andersen, Paul Bürkner, Akash Dakar, Alejandro Catalina, Kunal Ghosh, Joona Karjalainen, Juho Kokkala, Måns Magnusson, Janne Ojanen, Topi Paananen, Markus Paasiniemi, Juho Piironen, Jaakko Riihimäki, Eero Siivola, Tuomas Sivula, Teemu Säilynoja, Jarno Vanhatalo.
The web page has been made with rmarkdown’s site generator.


Posted by uniqueone
,

안녕하세요, SLAM 공부중인 장형기입니다.

이번에 Visual-SLAM에서 필수적으로 사용되는 RANSAC, 그리고 에러 데이터를 (outlier) 제거하는데에 사용되는 M-Estimator 기술들을 설명하는 글을 적고 있습니다.

각각의 기술들의 기술적 진화와 OpenCV에서 구현한 방법및 문제점, 간단한 코드 구현, 그리고 산업/연구 계열에서 사용하는 최신 SOTA 방법론까지 커버해보고 있습니다.

현재 3개의 글이 더 추가될 예정이며, M-estimator의 개념과 사용법에 대해 추가할 예정입니다.

많은 관심 부탁드립니다. 감사합니다!

1) Outlier란?
https://cv-learn.com/Outlier-Robust-estimation-0-Outlier-fa0c9cb92fdc4de79900118766c55cbf

2) RANSAC 소개
https://cv-learn.com/Outlier-Robust-Estimation-1-RANSAC-82e8aefaa56e493cb324ad752f9242f6

3) 기존 RANSAC의 단점 + PROSAC 소개
https://cv-learn.com/Outlier-Robust-Estimation-2-RANSAC-PROSAC-c7243d860b454493b68ff8a34bff4deb

4) OpenCV RANSAC을 그냥 쓰면 안되는 이유 + 업그레이드 RANSAC
https://cv-learn.com/Outlier-Robust-Estimation-3-OpenCV-RANSAC-RANSAC-086b2977b5b14212a3be566861fa3647

Posted by uniqueone
,

1주 1논문 리뷰 프로젝트 공유
안녕하세요. TFKR! 대학 여름방학 마지막 주를 맞아, 제가 이번 방학동안 진행한 1주 1논문 리뷰 프로젝트를 공유하고자 합니다!
여름방학 8주간 5편의 딥러닝 관련 논문을 리뷰하는 영상을 만들어 유튜브에 업로드하였습니다. 영상 시청보다 자료를 선호하시는 분들을 위한 PDF 자료도 준비되어 있습니다!
영상: https://www.youtube.com/playlist?list=PLDU04UBu69OG6mJ8m5osSxxQ0CFSej_e1
PDF: https://github.com/skyil7/paperReview
회고: https://skyil.tistory.com/93
논문 선정, 리뷰 과정 등에서 항상 TFKR 분들께 많은 도움을 받고 있습니다! 제 자료가 좋은 선순환을 만드는데 도움이 되었으면 좋겠네요😊
TFKR 여러분 남은 여름 건강하고 알차게 보내시길 바랍니다! 감사합니다~

Posted by uniqueone
,

안녕하세요. 모두연 여러분.
얼마전에 열렸던 AI Grand Challenge 4차 대회의 경량화 트랙에서 사용했던 구현체를 공개합니다 (최종결과 우승, 팀 제이마플).
https://github.com/j-marple-dev/model_compression
저장소의 내용은 크게 Pruning / Knowledge Distillation / Quantization으로 구성되어 있으며, 그 외에도 경량화 단계에서 성능을 유지하는데 도움이 되는 학습방법들이 포함되어 었습니다.
더불어 초심자 단계에서부터 대회 참가까지 2달여 간의 준비과정 및 참고자료를 슬라이드로 제작해 보았습니다. 부디 누군가에게 도움이 된다면 좋겠습니다.
https://www.slideshare.net/KimJunghoon5/2020-dld-238284732
참고로 위 슬라이드의 내용은 AI:Festival(http://aifestival.hellodd.com/) 발표로도 소개될 예정입니다 (9월 2일). 추후 발표영상도 공유드리도록 하겠습니다.
------------------------------------------------------
내용
Network Architecture
MixNet
Fast DenseNet
Augmentation
AutoAugment
RandAugment
CutMix
CutOut
Loss
Cross Entropy Loss + Label Smoothing
Hinton Loss
LR Scheduler
Cosine Annealing + Initial Warmup
Unstructured Pruning
Lottery Ticket Hypothesis
Weight Rewinding
Learning Rate Rewinding
Structured Pruning
Magnitude Pruning
Network Slimming
Magnitude Pruning + Slimming
Channel-Wise Pruned Model Shrinker (Experimental)
Quantization
Post Training Quantization
Quantization Aware Training

Posted by uniqueone
,

A novel neural network to generate high-resolution images
For project and code/API/expert requests: https://catalyzex.com/paper/arxiv:2008.10399

Posted by uniqueone
,

https://www.facebook.com/groups/TensorFlowKR/permalink/1282565068751215/

Facebook 그룹

TensorFlow KR에 멤버 51,082명이 있습니다. TensorFlow (TF), 딥러닝의 모든 이야기를 나누는 곳, 텐서플로우 코리아(TF-KR)입니다. 잡담방: tensorflowkr.slack.com - 잡담방에 참여하고 싶으신 분은 https://tensorflow

www.facebook.com

안녕하세요, 왕초보입니다.
MacOS, Windows 또는 LInux ?
인공지능을 제대로 공부하려는데요, 어떤 운영체제에서 시작하는 것이
성능 측면에서 (동일한 하드웨어 가정)
커뮤니티 측면에서 (추후 Q&A 시)
바람직할지 여쭤봅니다. 미리 감사드립니다. ^____^
MacOS, Windows는 친숙하고, Linux는 필요시 더 친숙해지려구요.

리눅스 컴퓨터로 하시는걸 적극*10000 추천드립니다... 윈도우도 cmd가 있지만 터미널이 진짜 편하고 온라인으로 나와있는 튜토리얼들도 왠만해서는 리눅스에서 돌리기 편합니다...

개인적으로 제일 좋았던건, 리눅스 컴퓨터 좋은거 맞추고 맥으로 그 컴퓨터를 ssh해서 학습 돌리는 거 였습니다. 개발은 mac에서 하고, 학습은 linux으로 하는 식으루요

Linux가 성능이나, 커뮤니티 측면에서 여러모로 편리합니다.
Mac으로 ssh 접속도 편리하고요.

Mac이 익숙하시니, Linux에 큰 이질감 없이 금방 적응하시리라 믿습니다.









Posted by uniqueone
,

안녕하세요. ECCV 메인 컨퍼런스도 이제 이틀차에 접어 들었습니다.

어제 오프닝에서 다른 학회들과 마찬가지로 각종 시상식 코너가 있었는데, PAMI Everingham Prize를 받은 Johannes Schönberger가 눈에 확들어 와서 한번 소개를 드려도 좋겠다 싶어서.. 학회 쉬는 시간을 이용하여 또 글을 써봅니다.

이분의 대표 작품은 바로 COLMAP이라는 오픈소스 SfM (Structure From Motion) 소프트웨어 인데요,

여기서 SfM이란? 우리가 흔히 쓰는 카메라를 단안 카메라 (Monocular) 라고 부르는데요, 우리가 한쪽 눈을 가렸을 때 원근감이 없어지는 것처럼 카메라 하나만 가지고는 3차원 공간상의 구조 정보를 얻을 수가 없습니다. 하지만, 카메라를 이동 시키면 같은 장면을 여러각도에서 보게 되는 효과를 얻게 되죠.. 한쪽 눈 감고 좌우로 움직이면 가까이 있는 물체는 많이 움직이고 멀리있는 물체는 조금만 움직이는 것 처럼.. 카메라를 움직이면서 각 픽셀들이 어찌 이동하는 지를 이용하여 깊이 정보 (Disparity map또는 depth map)을 얻어내고 그걸 계속 차곡차곡 잘 합쳐서 공간이나 물체를 3D로 만들어 내는 기술입니다.

다시 본론으로 들어와서.. COLMAP은 그런 SfM방법 중 가장 오픈소스로서 문서화도 잘 되어있으면서 성능도 좋은 녀석이라, 요즘 2D 이미지로 3D 로 뭔가를 하고자하는 수 많은 연구에서 베이스라인으로도 쓰이고, weak-supervision이나 self-supervision값을 만들어 내는 용도로 활발하게 사용이 되고 있습니다. 궁금하신 분은 핸드폰으로 자기 책상을 여러각도로 찍어보고 프로그램을 돌려보면 오.. 이런게 있구나.. 하실 수도 있을 것 같네요 ㅎㅎ (https://github.com/colmap/colmap) 바이너리도 배포가 되고 있고, 리눅스(우분투)에 익숙하시다면 컴파일 난이도가 높지는 않습니다.

비전 분야에선 워낙 유명하신 ETH의 Marc Pollefeys의 제자로 교수님과 마찬가지로 현재는 MS에서 일을 하시고 있네요. 이분 박사 학위 논문을 읽어보면.. 정말 최신 SfM의 역사가 담겨있는 교과서 느낌이 들었는데.. 연구나 프로그래밍 실력이 정말 대단한 것 같음을 느꼈습니다.

혹시 로봇에 단안 카메라를 장착하고 이것저것 해보고 계시다면, SfM을 이용해서 로봇의 인지능력을 3D공간으로 끌어 올리시는 것도 좋은 시도가 되지 않을까 생각합니다.

P.S: 앞으로도 이렇게 과학계 샛별같은 사람 또는 대가님들을 소개하는 글을 종종 올릴까하는데 그룹 분들의 의견도 함 들어보고 싶네요!

Posted by uniqueone
,


Pose2Mesh: Graph Convolutional Network for 3D Human Pose and Mesh Recovery from a 2D Human Pose 이번에 ECCV2020에 붙은 제 논문입니다.
in-the-wild image에서 3D human pose and mesh를 recover할 때 사용되는 3D 학습 데이터는 보통 controlled environment (ex. 실내 스튜디오)에서 얻어지는데, 이 때 in-the-wild 데이터와 controlled environments 데이터의 image appearance 차이로 인한 domain gap 문제를 해결하고자 하는 논문입니다.
관심있으신 분은 youtube 영상 시청해주세요!
youtube: https://www.youtube.com/watch?v=utaHeByNauc&t=12s
github: https://github.com/hongsukchoi/Pose2Mesh_RELEASE
arxiv: https://arxiv.org/abs/2008.09047


Posted by uniqueone
,

Super Fast and Accurate 3D Object Detection based on LiDAR
Fast training, Fast inference
An Anchor-free approach
No Non-Max-Suppression

Model:
ResNet-based Keypoint Feature Pyramid Network (KFPN)

Inputs: Bird-eye-view (BEV) maps that are encoded by height, intensity, and density of 3D LiDAR point clouds.

Outputs: 7 degrees of freedom (7-DOF) of objects: (cx, cy, cz, l, w, h, θ)
cx, cy, cz: The 3D center objects.
l, w, h: length, width, and height of the bounding box.
θ: The heading angle in radians of the bounding box.
Objects: Cars, Pedestrians, Cyclists.

The pre-trained model has been released in the repo.

Source code: https://github.com/maudzung/Super-Fast-Accurate-3D-Object-Detection

Posted by uniqueone
,

간단히 바로 가져다 쓸 수 있는 easyOCR API 소개 드립니다.

강원도 놀러갈 때마다 보이는 간판인데, 신나기 시작하는 "인제 - 신남" 간판 한번 테스트 해보았습니다.

아래 처럼 POST로 바로 호출할 수 있습니다.

curl -X POST "https://master-easy-ocr-wook-2.endpoint.ainize.ai/word_extraction" -H "accept: images/*" -H "Content-Type: multipart/form-data" -F "language=ko" -F "base_image=@인제신남.jpg;type=image/jpeg"

github: https://github.com/JaidedAI/EasyOCR

demo:

https://master-easy-ocr-wook-2.endpoint.ainize.ai/

[https://github.com/Wook-2/EasyOCR](https://github.com/Wook-2/EasyOCR)

Posted by uniqueone
,

SNU AI (구 SNU TF) 스터디 모임 7기에서 매주 발표되는 논문리뷰 자료들을 정리하여 TF Korea 분들께 카드뉴스형식으로 정리한 SNACK U AI Week 마지막주차!🧑🏽‍💻👩🏾‍💻를 공유드립니다!
마지막까지 SNU AI 성격답게 병리학부터 게임이론까지! 다양한 논문이 나왔네요!
다들 월요팅하시고 월요일 맞이로 논문 읽기도, 쓰기도 하는 SNU AI 8기 신청은 어떠신가요~? 🤷‍♀️🤷‍♀️ (온라인 예정이지만, 코로나 진행상황에 따라 차 후 온/오프 병행예정입니다 :) )
* 신청 링크: https://www.facebook.com/groups/TensorFlowKR/permalink/1276550572685998
Week 10 part 1(안재만): Hidden Technical Debt in Machine Learning Systems
https://papers.nips.cc/paper/5656-hidden-technical-debt-in-machine-learning-systems.pdf
Week 10 part 2(이홍석): HoVer-Net
https://arxiv.org/abs/1812.06499
Week 11 part 1(정욱재) : DynaBERT: Dynamic BERT with Adaptive Width and Depth
Week 11 part 2(Seo Taek Kong): Regret Minimization in Games with Incomplete Information
https://papers.nips.cc/paper/3306-regret-minimization-in-games-with-incomplete-information
Optimistic Regret Minimization for Extensive-Form Games via Dilated Distance-Generating Functions
http://papers.neurips.cc/paper/8764-optimistic-regret-minimization-for-extensive-form-games-via-dilated-distance-generating-functions
자세한 자료는 7기 발표자료를 모아둔https://drive.google.com/drive/folders/0B8z5oUpB2DysbFNEOWxfVDh5VW8에서 확인하실 수 있습니다!
Instagram: instagram.com/snu_ai_official

Posted by uniqueone
,

안녕하세요. 새로 운영진에 합류하게 된, 그리고 비엔나에서 로봇 비전 연구하는 박기루 입니다. 운영진이 되고나서.. 첫글로 제 논문 홍보를 하기는 좀 그래서 방금 끝난 ECCV 워크샵 중 가장 로봇과 가까운 분야 중 하나라고 할 수 있는 Recovering 6D Object Pose 워크샵에서 열린 BOP Challenge소식도 함께 준비해 보았습니다.

저의 연구주제인 Object Pose estimation관련해서 SOTA방법론들과 전통방법의 대표인 PPF(Point Pair Feature)가 경쟁하는 BOP challenge가 작년에 이어서 Recovering 6D Object Pose workshop (http://cmp.felk.cvut.cz/sixd/workshop_2020/) 과 함께 열렸는데요, 작년에 비해 훨씬 뜨거웠던 경쟁 끝에 드디어 2회 우승에 빛나는 PPF를 물리치고 CosyPose라는 방법이 1등을 차지하였습니다. 이번 ECCV에서 발표 예정인 이 방법은 엄청 Extreme한 Data augmentation을 적용한게 매우 인상깊은 방법이었습니다. RGB이미지만 이용하다보니 Depth이미지를 이용한 ICP(Iterative Closest Point)를 사용하지 않았다면 또 다시 PPF에 1등 자리를 내주는 상황이었는데 ICP덕에 1등을 하게 되었네요. (저자가 ICP는 제 Pix2Pose코드를 사용하였다는..깨알 자랑) RGB만 가지고 얻은 결과도 정말 박수가 나올만큼 경이로운 기록을 세우긴 하였습니다. 이제 CNN을 이용한 Pose Estimation연구도 점점 정형화가 되고 자리가 잡아가는 느낌을 받고, 더 좋은 트레이닝을 위한 렌더링 방법들도 좋아지고 있는 것 같습니다. 아쉽게도 이제는 퇴물(?)인.. 저의 Pix2Pose는 4위에 그쳤고.. 작년 부터 유지해오던 YCB-V 데이터 1위자리도 결국 내어주고 말았습니다. 1년 사이에 새로나온 논문들의 성능향상이 너무 눈부셔서 놀라움의 연속이었습니다.

아무튼 더 본론으로 들어가서, BOP Challenge에 사용 된 데이터셋들은 모두 3D 모델을 제공과 더불어 모델에 Texture까지 깔끔하게 입혀져 있어서 렌더링에 사용이 되고, 몇몇 데이터셋은 수천장의 Real image에 Pose를 일일이 체크해서 학습용으로 제공을 합니다. 하지만, dataset에 있는 물체가 아닌 내 책상에 올려져 있는 실제 물건들의 Pose를 측정하려면 어떻게 해야할까요? 모델에 Texture도 없고.. 수백, 수천장의 이미지에 Pose를 annotation할 시간도, 인력도 없다면? 바로 이런 상황에서 CNN들을 학습 시킬 수 있는 이미지를 만들어내는 연구를 이번 ECCV에서 Spotlight로 발표하게 되었습니다. ECCV 참석 중인 분들은 월요일(8월 24일) 오전 6시(한국 오후2시), 2시 (한국 오후10시) Live Q&A Session으로 초대드립니다. 최첨단(?) 크로마키 효과가 들어간 10분발표도 많이 시청해주시고요. ECCV참석과는 별개로 비슷한 연구 주제를 하고 계시거나 자세한 설명, 토론이 필요하시면 메신저로 연락주시고 즐겁게 이야기 나누면 좋을 것 같아요!

논문:https://arxiv.org/abs/2005.03717

코드:https://github.com/kirumang/NOL

Dataset:https://www.acin.tuwien.ac.at/en/vision-for-robotics/software-tools/smot/

ECCV paper site link (등록 하신 분만 보실 수 있는 링크): https://papers.eccv2020.eu/paper/2636/

Video: https://www.youtube.com/watch?v=fQJPS01cmac&t=9s

유럽, 한국 모두 코로나가 다시 심해지고 있는 이시기,

모두 건강하시고, 즐거운 ECCV 되시길 바랍니다!

Posted by uniqueone
,

안녕하세요 lidar slam 공부 중인 기섭입니다.

lidar slam 과 딥러닝을 어떻게 접목해야 할지 평소에 고민이 많은데요,
다름이 아니라 3D DL 동향에 대해 간단히 소개해드리려고 합니다.

robotics 에서 주로 lidar 를 통해 3D point cloud 를 얻게 됩니다.
하지만 3D data 는 image 와 달리 pixel구조를 갖지 않기 때문에 ("unstructured" and "unordered") visual domain 에서 잘 개발된 2D convolution 및 노하우들을 사용하기 어려운 단점이 있었습니다.

1. 시즌 1 - 시작 : PointNet 계열
그래서 PointNet (17 CVPR)이 point를 MLP를 통해 학습하면 이 문제를 해결할 수 있다고 보였고, 이후 연구들은 좀 더 receptive filed 를 어떻게 하면 더 flexible 하게 가져갈 수 있을지, 그래서 local structure 를 더 잘 포착해서 더 좋은 피처를 만들 수 있을지에 관해 연구들이 되어온거같습니다.
주요논문: PointNet, PointNet++, DGCNN, KPConv

2. 로보틱스 - 우회? : projection and 2D conv 계열
하지만 왠지 로보틱스에서 접하는 large-scale and sparse lidar scan 의 경우 pointnet 기반의 방법이 잘 안된다는 심증 + 물증(19 ICCV SemanticKITTI가 지적한 바 있습니다) 들이 학계에 있는거 같구요,
그래서 2D range image 상에서 하면 잘된다는 방법들이 주로 개발되어 오고 있는거같습니다.
(RangeNet++, LO-Net, DeepLO)
최근에는 Ouster lidar 같이 2D form으로 잘 arranged 된 lidar 도 개발되고 있어서 역시 이런 우회?도 타당하고 분명히 효과있는 것 같구요.

3. 시즌 2? : Sparse Conv 계열
근데 2D projection image 가 과연 최선일까 라고 계속 생각이 들긴 하는데요,
최근에는 3D conv 를 MLP로 우회해서 하는 것이 아니라 기존의 conv 연산을 sparse 하게 할 수 있는 툴들이 공개되고 있는거같습니다. 그래서 3D or 더 높은 k-dim 에서 연산을 바로 할 수 있는 거 같구요.
주요논문: SparseConvNet (18 CVPR), MinkowskiNet (19 CVPR)
저도 아직 방법은 자세히 이해하지는 못했는데요,
최근 (2019, 2020) 에 급격히 이 계열의 툴 (Minkowski Engine) 을 이용한 applications 들이 등장하고 잘된다고 보고가 되고 있는 것 같습니다. https://github.com/NVIDIA/MinkowskiEngine/wiki/Usage
타겟 어플리케이션의 종류도 segmentation 등에 국한되지 않은 것 같구요

그래서 결국 최신의 3D DL tool 을 어떻게 lidar slam 과 물릴지 관심이 많은데요
관심만 많습니다..
비슷한 고민있으신 분들 종종 이야기나눌 기회가 있다면 좋을 거 같습니다.

감사합니다!



Posted by uniqueone
,