The course lectures are solid, but the assignments are pretty dismal for beginners. There isn't much guidance built into the assignments, and sometimes they require the use of things that were absolutely not covered in the lectures(classic academic mistake). My suggestion for the course creators is to examine how Andrew Ng's assignments are in his Coursera course and model them after that. Or simply make sure that the assignments are clear(clear to someone beginning, not a TF expert).

I liked the lectures (videos). And I did not like that the course has no mandatory programming assignments. I pay for the course to make myself study. And I believe that there is no study without practice. Hence, this course did not make me study, thus I don't understand why I need this course :-(. And I could find free lectures about TF/Keras (maybe not so good, but free) and/or read the documentation. BTW, I really like Andrew NG's courses, but this one really disappointed me.

This course focuses on the teaching of TensorFlow modules related to CNNs and does a good job in introducing some modules of tf and keras for data loading and manipulation. However, it is very light on theory and is only helpful if Deep learning specialization is taken beforehand or in conjunction. Furthermore, this course will need some refresh soon for its modules as it is still using version v1.x of tf as well as some code re-organization.

The use of the .flow() method on the last exercise would deserve some explanations : the labels need to be transformed from sparse (int format) to one_hot (with tf.keras.utils.to_categorical for example), so the loss='categorical_crossentropy' actually works in model.compile().

There is no mention of the different ways of structuring the labels in the course, this can be misleading.

Other than that, good material.

The videos were quite solid. The programming assignments were poorly designed to accept identical answers, but not other solutions that work. This did not evaluate students' creativity and depth of understanding. The Jupyter notebook environment was quite fragile. The resources allocated for each notebook was quite limited. I expect more computer or human resources allocated for each student paying the tuition.

The course generally was fine and it taught me many things on how to use tensorflow for aumentation and regularization. But...I think that notebooks need a little bit more clarification. Many times I didnt know exactly what to do and other times the comments were misleading. Overall I would recommend to get the free trial and see if you like it before you spend the 40-50$

This course teaches you how to apply CNN to image data, how to augment image data with ImageDataGenerator, and how to do transfer learning. It is very easy to follow, and quite possible to finish in half a days worth of effort. It would be nice to be more explicit with what is required by the grader, as assignment instructions not always are clear.

Some Excercises are not great:

- Requirements and goals are not set in the outset

- Functions need to be used that were not explained, nor even hinted at

- Sometimes significant focus of an exercise is put onto modifying file structure and/or csv importing (which is not what I want to spend my time on in a Tensorflow Specialization)

Good but too basic.

Specialization's first course already covered the basic of tensorlfow. This course is suppose to expose to sota topics in computer vision using cnns. The content in this course can be easily fetched from many online forums. Thus the curators need to put some advance topic like attention, spatial transformer etc etc

The videos and explanations by Laurence and Andrew are good, but I did not like the programming assignments in this course, because of their lack of explanation 'what to do'.

The programming assignments really need some fixing. They are not to difficult, but they lack explanation of what to do, which parameters to use, etc.

The deep learning concept in this course was to the point but the only thing i didn't like is the preprocessing phase. Like the students are not taught how to preprocess the data what i mean by that is that we are not taught how to create new directories and how to place images in those directories and things like that.

A good course, but again, not nearly as in depth as the original deeplearning.ai set of classes. The material feels introductory and at times superficial, with no real work required of the student to complete the class. At best a very early start to using convolutional networks with the keras apis in tensorflow.

I am giving only 3 stars because of two reasons: 1)the content is not significantly different than course 1. I didn't feel that I learned a lot more than course 1.

2)Assignment for week 4 is not well structured. Instructions are not clear. Moreover grader is poor quality and keeps running out of memory.

Not much recommended! Leave out too many details both theoretically and technically. The quizzes and the coding assignments are not well-designed. Specifically, the expressions in the quizzes are kind of sloppy and the coding sometimes requires tedious and repeated (no more than copy and paste) work.

I learnt a lot about CNNs and how to implement them, but I was taken aback to see advanced coding concepts being used in the programming assignments. I thought the concepts taught in the course itself were to be used only, but some parts of the assignments had parts which were too much to grasp well.

The course was very repetitive, not challenging, and therefore not particularly helpful. Andrew Ng's Deep Learning Specialization is vastly superior. Aside from getting used to TF and CoLab, I'm not sure what this helps with. I found it odd that it was recommended to me after the DL specialization.

Videos are great, but a little bit short. Comparing to AndrewNG courses and slides, the videos are merely the trailer for course. Grading is not what I would be expecting and it is one of worst I have seen on Coursera related to AI/ML. I was expecting a little bit more from this course.

This course is a really light introduction with CNNs in TensorFlow. While I enjoyed the videos, the content feels far too shallow. I completed the course in a couple days (and I'm not an expert in the field). It felt more like having gone through a TF tutorial than a grad-level MOOC

It is a good course for a fast overview on this topic. Be aware that it is not an introduction on ConvNN (but there are several courses of deeplearning.ai on this topic). If you are looking for a detailed course on Tf for ConvNN, I suggest you a book, the official documentation.

Basically a shallow introduction to programming simple CNNs with Keras. A lot is reused from the first course in the specialization. Reading one of the Tensorflow Tutorials/API documents on CNNs, Dropout, and TransferLearning will be time better spend, than doing this course.

The course is Ok, it gives you some insight on CNN and some useful tools in the Keras API. However it is quite simple and it doesn't explain the fundamentals behind it. The final tests are very simple, but can get quite complicated if you don't attached yourself to the tips.

The course itself is really good; however, homework problems at the end of the chapters are very unorganized. There is almost no guide at all. You may end up spending hours while trying to figure out why grader is having problems or your model's accuracy is very low.

Giving it 3 stars because of the last week's assignment. There is little to no information about the dataset and the learner is just expected to know how to deal with the data. No information on how many classes to expect as output and other necessary information.

What could improve it: Not enough depth in the practicals if you have already done Andrew Ng's course on Conv nets. No graded practical exercise.

What was good: Clear examples, Good setup to experiment with the algorithms & Speak explains concepts very clearly,

Good course, but the notebooks need a deep review to fix the problems related to balance between the requirements of the exercise and the resources available also a better explanation of the exercise aims would be a nice to have to avoid misunderstandings