judasdie / deeplearning.ai Goto Github PK

Some work of Andrew Ng's course on Coursera

Jupyter Notebook 99.07% Python 0.93%

deeplearning.ai's Introduction

deepleraning.ai-course

This is some records of Andrews' deeplearning.ai course from Coursera. I have uploaded my programming implement to this warehouse. There might be something wrong with my work, be free to email [email protected]. And if you can push your right answer, that will be grateful.
And to be clear, Ng's course is my first access to AI. I would love to share more learning experience on DL/ML, so I will put some good repos/articles on this page. Please read this whole REDEME, hope there is some thing helps you.
Read AI article on my blog

Course Contents

Neural Networks and Deep Learning
Week1　Introduction to deep learning
Week2　Neural Networks Basics
Week3　Shallow Neural networks
Week4　Deep Neural Networks

Improving Deep Neural Networks
Week1　Practical aspects of Deep Learning(Initialization-Regularization-Gradient Checking)
Week2　Optimization algorithms
Week3　Hyperparameter tuning, Batch Normalization and Programming Frameworks

Convolutional Neural Network
Week1　Foundations of Convolutional Neural Networks
Week2　Deep convolutional models: case studies
Week3　Object detection
Week4　Special applications: Face recognition & Neural style transfer

Sequence Models
Week1　Recurrent Neural Networks
Week2　Natural Language Processing & Word Embeddings
Week3　Sequence models & Attention mechanism

Blog Articles

About install tensorflow-gpu on windows

Cause I found many people get many issues while installing tensorflow-gpu on windows, and there are many talks before tensorflow-1.4.0 shows like this on Stackoverflow.
You will not face problem above again thanks to tensorflow-1.4.0!
python3.6.x + cudnn6.0 + CUDA8.0 + tensorflow-1.4.0 work perfect!
Remember move files in cundnn to CUDA8.0 accordingly (you also do this in the past)!

Learn Tensorflow and Deep Neural Network

I recommend you a video course for learning tensorflow from Google here
A good introduction about Deep Neural Network, download here
Best results on standard dataset like MNIST, CIFAR-10/100, ILSVRC2012... here
Keras Documentation Chinese Version here
Deep Learning by Goodfellow here

Some Good Machine Learning Tutorial

Expectation Maximization(EM) course by Xu Yida on Youtube
Understanding Variational Autoencoders (VAEs) from two perspectives: deep learning and graphical models

Some good repo

GANs with keras here
GANs with pytorch here

Update Records

Specialization 4 Done!　2017/11/13
Move Course Records to Records.md　2017/11/14
Add Expectation Maximization(EM) course links　2017/12/13
Add a Chinese Version keras tutorial　2018/2/23
Specialization 5 Done!　2018/3/8
Add GANs repo 2018/5/13

Thanks for Andrew Ng!

deeplearning.ai's People

Contributors

Stargazers

Watchers

Forkers

ghellstern pushkargole xinyuecai2016 zhejia vamsi53 xyp8023 alomdaelmasry janismdhanbad alfredo203 weacera ionuttamas yu-shengfeng yowhatever fengzean earthaa otherprojectsforks vanducdoan xingminglei xiaojinwhu naqute yq277 hosinprime yyhaoshuang nile649 yjhbnb jsmaxim bigzhao drah-kah-ris zhuohuwu0603 obinsc boonken13 kinzlp bigrlab miranthajayatilake wandresvr muxare boogh immuntasir littlerabbithole kickreg dalabs nguyendinhhuynh gokulramanaa techeye220 alexandrudaia daiaatena paapu88 jablkoadj ricardozitseng kevinwkc zzzhacker ramchak1 obitoquilt adndcn amansrivastava17 peterjiangwy zju-plp mgao3000 atenaalexie bajajtushar meelement saturdaysai james-fu angleboy8 lxianwei003 ngupta1137 nsahaytech yshenuoe raka747 novav barbecacov thtrnh tninh byrantwithyou 120674085 manghat jianxinzhou locussam chinmayi5577 solomonyuan bananawil taeng999 aijianiula0601 sergiojx sivunq starry91 manoj652 georgeginis ajaysimhasr mmelodious codingmylife phdkiran 9hy xingchen-yu jackyfriend swarupg1 shanhaoyu chisyliu bingai sam186

deeplearning.ai's Issues

Week 1 Error

Your error in Week 1 is the vert_start should be h * stride.

not have images folders

images/RNN.png so and on ,can not find

Error in Week4 / NST

When computing the triplet_loss your function outputs 350 while it should output 528

I get a error message when I run yad2k.py

inception_blocks_v2.py

When I try to run this code, it says ValueError: A Concatenate layer requires inputs with matching shapes except for the concat axis. Got inputs shapes: [(None, 128, 12, 192), (None, 32, 12, 192), (None, 32, 12, 102), (None, 64, 12, 192)]. Would you please help me in this issues? It will be a great help. Thank you.

YOLO algorithm on Videos

In week 3 content of CNN's course, we applied YOLO algorithm for predicting bounding boxes on images. How can we apply the same for videos.

Convert .h5(Yolov5) file to .pb file

I can successfully train my own dataset and get a model.h5 file after training. However, how do I convert this .h5 file to .pb file that I can connect the trained model with Labview. Thanks.

Sequence Model

When you update fully sequence model? thank you

Some answers are not correct

I see some answers are not correct

Where can I get the yolo.h5 file?

Great content. I was following the car detection notebook in the Week3 of Convolutional Neural Networks. I notices that the pretrained yolo.h5 file is missing from the directory. Is there any place I can download this from?

Thanks in advance.

Error in Week 3/ Object Detection

In the calculation of IOU, it is not correct for the cases where there is no intersection between the two anchor boxes.
Ideally it should return 0

How to load the pretrained weights file for FaceNet?

I'm trying out Facenet code locally, and having difficulties loading the pretrained model.
It's giving an error in the load_weights function in the fr_utils.py as below;
conv_w = genfromtxt(paths[name + '_w'], delimiter=',', dtype=None) KeyError: 'conv1_w'

Can you help to sort this out?
Thanks.

Question about Face Verification

Dear @JudasDie,
Thank you for your fantastic repository. I have a question about your Face Verification system (week4, Section 3.1). Your designed face verification system requires a predefined threshold (e.g., 0.7). Choosing an appropriate threshold value would be an annoying issue. In order to design a face verification system which doesn't require threshold, and clearly defines the same/different faces status of two input face images, I think it is better to design a new system such as below (i.e., In fact, I want to design a Siamese Face Verification system which produces the confidence score of two input faces, which belong to the same face or different one):
1- By using the FaceNet pre-trained weights one can extract the 128-dimensional embedding of two input faces.
2- Compute the L2 distance (or maybe L1 distance) of these two embeddings.
3- Pass the above-mentioned 128 vector to a Fully connected layer (128->2) followed by a Sigmoid (In order to get the confidence score of each class (i.e., Same Face class or Different Face class)).
and we just need to train the last mentioned Fully Connected layer.
I want to ask you that does this mentioned solution correct? What's your opinion?
I would really appropriate if you can help me.

In generate_colors why are we multiplying by 255 after converting to rgb?

Error in Week 4 / NST

When computing the J_style_layer your function outputs 11 while it should output 9.19028.