The lectures are really good and quite engaging. The extra course content by Dr. Ng is also generally where I learn the most. This class does a decent job in introducing to you the Tensorflow library.

However...

It feels a bit like an very well done tutorial. After finishing my second class I don't any any more idea of how to build a neural network than I did before. The data that they give you has already been cleaned and is ready to use which never happens in real life. The data manipulations that they ask you to do in the homework has zero explanation of why you are doing it. Just that add an extra dimension or split the data this way. I don't know why we need it split that way and it never says why. Further, exercise 4 in particular uses a different method than has been used for the entirety of first and second class. You're given a list of numbers rather than real images.

At the end of the class I wanted to understand how to build these models in the real world. If I want to predict cats vs dogs then great! However, if I want to try to categorize financial transactions or predict fraud or literally anything else, this class gives you no understanding of where to start, or how to approach the problem generally.

The course does not provide much knowledge. In fact this is a tiny extension compared to first part ("Introduction to TensorFlow for..."). The assignments are trivial - you need to change one or two parameters from the excercise codes. But another story is that the codes you succesfully run in Jupyter need to be tailored to satisfy coursera system (e.g. data cannot be loaded due to out of memory, TF version is different etc.). Instead of working on the model, you loose time on experimenting with the code unrelated to tensorflow or look at the forum for solutions specically suited to pass the test (they do not change how the models works). Do not recommend this course.

The first and fourth graded assignments were not very well posed. The grader in the 4th graded assignment kept running out of memory. The instructors do not get back to people in the forums. There was not much actual new material: most of the 4 weeks of material could have been covered in a single week. This has been the most discouraging coursera course i have taken.

This specialization is false advertising. It does NOT prepare you for the Tensorflow certification exam. It’s especially disappointing after taking the fantastic specialization by Andrew Ng, and makes this specialization feel like a cheap cash grab by Coursera and DeepLearning.ai. This series of courses fails to prepare you for three reasons:

1 – The certification exam is done on whatever is the current version of Tensorflow (v2.6 as of writing). You can’t expect a specialization like this to update every minor release, but much of the coding is still on the v1.X version!

2 – The certification exam requires you to work in the PyCharm IDE. The IDE doesn’t even get a mention in this specialization and it is all done through Google Colab.

3 – The material is covered at a very superficial level. I was hoping to walk out of it feeling confident in using Tensorflow on novel problems, but I’ve barely learned anything about Tensorflow that I didn’t already get from Andrew Ng’s specialization. There’re a few minutes of lectures (some weeks less than 10 minutes). The programming assignments are either pathetically easy, or lack any guidance on what to do (seriously sometimes there’s no instructions at all, you have to guess what to do by the variable names), or both.

Save your time and money and go elsewhere to learn Tensorflow.

This is one of the worst courses I have taken in Coursera though I expected it would be one of the best. This is a nothing course actually. The only two things I learned from the course is implementing transfer learning (and how to call ImageDataGenrator). The instructor is more interested to show off his synthesized dataset rather than teaching anything real. There was so much to teach like how to implement inception module, batch normalization, regularization (though dropout was introduced), and many more. Instead half of the course he was busy showing images from his datasets. Coding assignments were buggy too, the environment often crushes. Also, the last two weeks' assignment are mostly data cleaning. Really disappointed. Andrew NG set a really high standard for deeplearning.ai, other courses should maintain that level.

the exercises are not very challenging and most of them are the same as the practice colab notebooks, there should be more variety. And also the grading in the exercises is not good there are a lot of errors and it should be made more robust to individual changes made by the student

Very shallow and full of typos! And the staff doesn't care.

It is ridiculous.

Studying Convolutional Neural Networks (CNNs) in TensorFlow has been a highly rewarding experience for me. The courses offered a comprehensive understanding of CNNs, covering the fundamental concepts and practical applications. The content was presented in a clear and concise manner, making it easy to follow along and grasp the core principles behind CNNs. Through the courses, I gained a solid understanding of convolutional layers, pooling operations, and how they are used in tasks such as image classification and object detection.

One aspect that impressed me was the hands-on approach taken throughout the courses. The practical exercises and coding examples provided invaluable opportunities to implement CNN models in TensorFlow. By working through these exercises, I not only solidified my understanding of the theory but also gained confidence in implementing CNNs and experimenting with different architectures and hyperparameters. The courses' emphasis on practical application allowed me to build a strong foundation in working with CNNs in TensorFlow.

To be fair. I actually struggled with the final programming assignment, last function, with some TensorFlow background and plenty of PyTorch background. Guess they took into consideration some of the negative feedbacks in the last 2 years and modified it the assignment difficulty.

Majority of the programming assignment should be relatively simple copy and paste if you know what you are doing.

For the last part you cant copy paste everything directly, neither was the code issue covered in the lecture sample code. In the sample code, the label uses one hot encoding vectors to do multi-classification, in the assignment, an numerical category classification label was used. Need to google around for solution.

I have rated the course 5 star for the instructors and topics covererd. Though for the last assignments there were many things which I found no reference until I went through the discussion forum and I felt more teaching assistants' involvement is required because many of the questions were answered by the students only and many were unanswered,it would be more like a good guide to us if we get comments from the teaching assistants themselves . But overall when the instructor is Mr.Laurence Moroney, there's no doubt that the learning journey would be completely fun and seems easy. Thank you sir and coursera.

This was a great course to go more in depth about the use and implementation of convolutional neural networks. Learn the concept and implementation of "transfer learning" or "inception" to take advantage of CNN trained over a much larger dataset and fine tune the DNN to specialize on a different (but smaller) problem. It was great the learn that one can drop out some neurons from the pre-trained CNN in order to avoid overfitting and specialization. Changing from binary classification problems to multi-class problem is very easy to implement in Tensor Flow / Keras.

Excellent course for a beginner like me. It definitely helped me gaining the concepts and insights of transfer learning and the multiclass classifiers. I am confident now in dealing with the convolution neural networks, coding them from the scratch and to achieve the desired accuracy. The concept of dropout layers has been conveyed in the best possible manner and its affect on the validation accuracy can be easily observed. I would like to appreciate the efforts of the team of Coursera and the instructors for laying down an extraordinary online lecture series.

Great for people who want to not delve too deep into theory and learn the latest tools to get going quickly. I had already done the Deep Learning specialization so I recommend that as a great complement for the theory part. I learned everything I needed to get going with a practical application in this course. My only complaint is that I felt that the quizzes were poorly designed - most questions emphasized whether you remembered a specific API's argument name, or some questions were a bit ambiguous. Otherwise, highly highly recommend the course.

This was just an exceptionally well-done course. It's not complicated, but I don't think the point of it is to be complicated, just practical. All in all, I enjoy the teacher's style. If you're trying to understand the fundamentals of the theory and mathematics, these courses aren't for you; if you're looking to just gain a practical and useful working knowledge, then this is a great starting place. I took it to just round out my understanding of Tensorflow via Keras; this was a great course for that.

First of all, the course was amazing! I found it great for the following reasons:

- Laurence Moroney (Instructor) was very professional and clear while delivering the knowledge

- The introductions by Andrew NG were really nice

- Easy to understand codes and understanding of thr underlying principles

- Varied topics such as CNN, NLP & Time Series

- Very insightful by providing expert opinions about different ways of model optimization

I really enjoyed the course and I thank the instructor for the same :)

I am already familiar with machine learing and convolutional neural networks, and before starting using the TensorFlow framework I wanted to develop my own know-how in order to really have control and knowledge on what am I doing. Now that my C++/CUDA implementations work, I feel allowed to use a better tool like TensorFlow / Keras, and I am really discovering their power and flexibility, and I am getting really excited of the productivity that I can gain in my projects thanks to them!

Laurence Moroney presents another superb primer on the mechanics of tensor flow. Heavy on image analysis, we see how convolutional nets — concatenating stages of convolutional filters and pooling — extract features from images at whatever scale they appear. The exercises contain a modicum of basic-python skills reinforcement. Upon completion, one is equipped to tackle other common problems, e.g., the usps handwritten-digits challenge https://www.kaggle.com/bistaumanga/usps-dataset.

The pace of these two courses and the extremely to-the-point nature of the explanations, examples, and exercises enabled me to implement customized CNN-based codes my own data in only 5 weeks. Now I am definitely more confident to explore and implement more complex models and concepts in Tensorflow. Thanks to Andrew, Laurence, and the rest of the team for the very efficient learning experience and for sharing their knowledge and expertise.

This course makes me have the sense of how does it feel like to design a network for a problem.

Without any guidance, it's difficult to have a right guess at the beginning.

Transfer learning might be helpful if your target is to apply some kind of developed network into your application.

And also, one has to practice to be able to use the Tensorflow framework fluently.

Because there are so much concepts and corresponding APIs.

Great course, great content, and the best part is you are getting quiz and those "Challenging, interesting, excellently" designed assignments which surely test and improve your real skills. I'm just excited about those assignments every time I progress with a week.

Till now I have completed the first two courses of this Specialization and I'm sure the rest would also be great. I would be taking them all!

greate introduction to Image Classification. The skills is very very useful!

I like this course.

My advise to other learners is reading keras official developer guide(https://keras.io/guides/) when you learn this course. That will be very useful.

Besides, I want to get more skills about Image Segmentaton, Object Detection ,etc. So I hope Deeplearning,ai launch more advanced Computer Vision courses.

It is really an amazing course, My heartfelt thanks to Mr.Laurence Moroney, for his great teaching and Mr. Andrew Ng for giving these great platform. I Really enjoyed the course. I learned it lot of things here. I am going to take all the specialization in these courses. And It is great pleasure to thank Coursera platform for providing me Financial aid to take up these course.

Thanks

Rishiganesh.V

Great course, well structured and straight to the point, the point being application. Can't recommend it enough for those who completed the deeplearning.ai specialization.

With no sufficient theoretical knowledge and simple python programming, however, the course is vague and highly not recommended. Sufficient understating of how DNN work greatly improves the added value of this course.

Great and extremely interesting.

Last assignment is tricky and probably corrupt, as it does not explicitly require additional label processing, which is necessary to successfully train the model. However, the grader is not "aware" of this additional processing and is lowering the grade, expecting the labels to be passed in the form which is not accepted by the model. Please fix that.