Deep Learning for Computer Vision with Python and TensorFlow – Complete Course

29 thoughts on “Deep Learning for Computer Vision with Python and TensorFlow – Complete Course”

• Awesome course and projects on CNN! Love your work. P.S. (8:29:15) Loading and Saving Models to Google Drive : This section is just still screen. There's no change.

• how many of you have completed the whole course?

• Day1 ;20 min

• Notes for myself:
  Day 1 2/7/24- 1:14:49
  Day 2 3/7/24 –5:00:00

• I want nlp

• I see lots of people putting their reading progress in the comments. To complete this course you will have to find the passion otherwise impossible (laziness because of the short formats). This is my 3rd time watching this course and I haven't quite grasped it all yet so don't expect to watch it just once.

• Watch at 1.6x

• Starting today, June 28th, 2024, 10:00 am ; ). I hope I make it to the end.

• FIRST TIME WITH DEEP LEARNING

  day1 – 2:56:21
  day2-nil
  day3 – 4:14:52
  day4-nil
  day5- 6:28:50
  day6- 8:47:10
  day7-10:00:00

• 22:07

• aint no way they released such a detailed course for free that's over a day!

• may God bless you

• I simply cant thank you guys enough the whole team for sharing such content to the world!. Where institutions charge u 10s if thousands of dollars for such degree . You are practically giving it away for free.
  Thank you so much everyone . I can only imagine the less fortunate people finding this on the net would be a game changer for them

• Thanks for the knowledge! I'm also starting to lean ML and DL, can anyone tell me how can I access the dataset for these models so I can deploy them on my own in VSCode? Thank you so much!

• so I was searching for a course like this, finally found one! I aim to complete it before july end, today is 30th may, will start from tomorrow and update each time I open this video (maybe this motivation will help me continue)
  god, I wasted my one month of vacation , let me not waste this too! fighting!!

• note for me
  Day 1, completed till 1:12:56
  Day 2, completed till 3:1:51

• Now I speak Chinese English.
  Thanks.

• If I watch full 37 hours, would I master computer vision?

• Thank you for this free course

• how i can download this video?

• RESNET_34 on augmented dataset or without augmented…..resnet_34 code and model details not clearly shown

• What's the matter with the video at 8:36:21? The teacher is talking as if he is executing some commands and nothing is shown except some static screen.

• Day 1: 1:07:31
  Day 2: 1:55:02
  Day 3: 4:54:47

• gotta put timestamps on the video itself rather than in the description

• I appreciate that this is for free, but it is so annoying the lack of practical examples in the first 4 hours. I know I suck at math but explaining the pure mathematical algorithms without involving any kind of real-life examples (like how these work with images) makes this so difficult.

• Great buddy

• I have a project at school related to Computer Vision which a field I have never learned. I WILL DECIDE TO COMPLETE THIS COURSE IN 2 WEEK.
  
  THIS COMMENT WAS WRITTEN TO SHARE MY EXPERIENCE SO THAT THOSE NEW TO THE VIDEO CAN WATCH IT FASTER.
  – If you have another laptop/PC or screen, you can watch and practise faster. Install microsoft mouse without borders in to workstation, if your workstation can't work with 2 screen. No one want ALT + TAB in 37 hours.
  
  – I think almost people decide practise with this video are not zero. If you are zero, you should watch another video to get some define and theory about machine learning and computer vison before watch this. Because you can learn deeper and faster when reviewing concepts a second time and practise better and you don't want to seeking back and forth this 37 hours video
  – Below the video description, there are work items. If you are determined, you should complete a item/subject/project before sleep. It will give you more motivation to survise these 37 hours of hell
  +) DAY 1: I don't learn much today
  ⌨ (0:00:00) Welcome: Skip or listen if you want
  
  ⌨ (0:05:54) Prerequisite: Skip or listen if you want
  
  ⌨ (0:06:11) What we shall Learn: Skip or listen if you want

  Tensors and Variables
  
  ⌨ (0:12:12) Basics : tf.constant, shape, ndim, dtype, cast, tf.zeros, tf.random
  
  ⌨ (0:19:26) Initialization and Casting: Nothing
  
  ⌨ (1:07:31) Indexing: All his code
  
  ⌨ (1:16:15) Maths Operations: tf.math.multi
  
  ⌨ (1:55:02) Linear Algebra Operations: tf.linalg.matmu, tf.transpose, tf.expand_dims
  
  ⌨ (2:56:21) Common TensorFlow Functions: learn all function
  
  ⌨ (3:50:15) Ragged Tensors: don't learn
  
  ⌨ (4:01:41) Sparse Tensors: learn
  
  ⌨ (4:04:23) String Tensors: don't learn.
  
  ⌨ (4:07:45) Variables: learn all

  – DON'T WATCH WASTE YOUR TIME if you already have a little knowledge of linear algebra, get the code of one in the description, combined with chatgpt you will automatically understand how to use tensorflow which he has achieved. This will save you a ton of time and actually the math is more detailed but don't break into the math because this is not linear algebra. If you still don't understand, look at the part you don't understand.
  
  – To me, these 4 hours of video are useless, it's really pointless to learn tensorflow because it is a library containing a lot of functions and you will only really need it when the project is relevant. And when problems occur, you will use chatgpt instead of the knowledge learned in these 4 hours. In my opinion, The above things are the most worth learning

  +) DAY 2: I have a big homework and can't learn today
  +) DAY 3:
  Building Neural Networks with TensorFlow [Car Price Prediction]
  
  ⌨ (4:14:52) Task Understanding
  
  ⌨ (4:19:47) Data Preparation
  
  ⌨ (4:54:47) Linear Regression Model
  
  ⌨ (5:10:18) Error Sanctioning
  
  ⌨ (5:24:53) Training and Optimization
  
  ⌨ (5:41:22) Performance Measurement
  
  ⌨ (5:44:18) Validation and Testing
  
  ⌨ (6:04:30) Corrective Measures
  Actually, I don't appreciate this project of his. He tried to introduce concepts to begginer, he integrated machine learning, linear regression, neutron network into one problem. The concepts of linear regression and how to train the model are explained quite well, but the neutron network is not very clear. I learned about neutron networks in college and it's actually a lot more detailed and easier to understand than here. This is an important topic and there is a lot of knowledge that needs to be imparted. It would be better if he taught the theory first and then practiced. The practice of both practicing and conveying is quite bad and has many shortcomings. In addition, dividing the dataset into 3 sets training_set, validation_set, test_set is usually done first but he puts it near the end, which upsets the logic.

  +) Day 4 + 5:

  Building Convolutional Neural Networks with TensorFlow [Malaria Diagnosis]
  
  ⌨ (6:28:50) Task Understanding
  
  ⌨ (6:37:40) Data Preparation
  
  ⌨ (6:57:40) Data Visualization: In his seaborn practise, visualization can give you information about dataset. You can detect more issue in your dataset and find orther method to solve
  
  ⌨ (7:00:20) Data Processing:
  
  ⌨ (7:08:50) How and Why ConvNets Work: Listen this carefully
  
  ⌨ (7:56:15) Building Convnets with TensorFlow
  
  ⌨ (8:02:39) Binary Crossentropy Loss
  
  ⌨ (8:10:15) Training Convnets
  
  ⌨ (8:23:33) Model Evaluation and Testing
  
  ⌨ (8:29:15) Loading and Saving Models to Google Drive: Colab is a virtual machine from Google and run linux os. You can save your model'architech (model layer -CNN/Dense/Pooling, loss function = RMSE, CrossEntropy, …. ) and model'weight after training. The worst thing in this section is it has only a image not video

  Overview this project is great for everyone start to learn about CNN like me
  Disadvantage: In fist section, he train model with 1 layer – 1 neutron to demonstrate linear regression. If he started with model has more dense layer, the project will be faster. Because the project is made to learn more theory, you must spent a little of time to complete this project

  My experience:
  – I was stuck at CNN concept a lot of time, I have read and watch more document to understand about it. This is my answer for everyone, don't understand after watch his CNN concept. In basic neutron, you have more input signal (X1, X2, …XN) and a bias X0. It will come neutron, a neutron has 2 thing: Aggregate function and Activation Function. With basic neutron, the Aggregate function is
  Net = f(X) = w0*X0 + w1* X1 + …. + wn*Xn
  – Training a model is find W = [w0, w1, … , wn] best fit with training set. If you choice that Aggrgate function, you must file a thousands parameter in training time with a neutron, and a huge parameter in model with more hidden layer, which has more netron in it.
  – You can image Convolution Layer is a fully-conected layer has more neutron. The Aggregate function offeach neutron is replaced by convolution calculus with tensor of image input and a kernel matrix has size (k*k*number of chanel of image). By the way, instead of find wi for each Xi input signal/bias, you just find the weight of kernel matrix => You just find (k*k*number of chanel of image + 1) parameter for each neutron. Finaly, each neutron in convolution layer is called "fliter" , the ouput of layer is called "feature map"
  – Google Colab limited time to use GPU, excecpt pay monney. When I train his model 2 times, I have reached limit. For everyone has a window laptop has NVIDIA GPU, install wsl 2 with Ubuntu distribution and install cuda+cdnn+clang+bazel in it. Note, you must install exactly version that fit with your tensorflow version (I have install tensorflow 2.16.1 with cland 17.0.2, bazel 6.5.0, cdnn 8..9 and cuda 12.3). Create a virrtual python environment 3.10 (if you install python 3.12 in it, you can't load dataset by his code. This is Dataset API of tensorflow 2.16.1 bug in the time I comment). Sea and following this video for sure https://www.youtube.com/watch?v=VE5OiQSfPLg&list=LL&index=1&t=889s&pp=gAQBiAQ , you can create virtual python environment with anaconda like this video or use vevn model in python3
  +) Day 6:

  Building More Advanced Models in Teno Convolutional Neural Networks with TensorFlow [Malaria Diagnosis]
  
  ⌨ (8:47:10) Functional API
  
  ⌨ (9:03:48) Model Subclassing
  
  ⌨ (9:19:05) Custom Layers

  First 2 parts, don't name function input, lenet_model, …. You will get errors when you call cell many time. In 9:18:26, the right is tf.zeros([1, IMSIZE, IMSIZE, 3]).

• Day 1: 42:55
  Day 2: 1:04:08

• Can anyone tell me if there is freelancing career in computer vision?

Comments are closed.