The content is generally great and well worth it. I wish they would fix some of the errors, especially in ungraded exercises. You end up wasting a lot of time because of them. Perhaps the most frustrating aspect is navigating to the Jupyter notebooks. I wish the links to the notebooks were on the same pages as the Submission and Discussion links.

It is a nice course but one bad thing in this course is the programming exercise, sometime it is not very clear or there are some places that function has changed but the function mentioned is the old version and another suggestion is i hope there are more vids on the syntax on tensorflow, one vid in course 2 is not enough to handle the assignment.

I liked the course and the topics were discussed and learned. I gain much more algorithms and tools for various ML topics.

However, I am still missing the tools to start a problem from scratch - i.e. gathering data, arranging the data, building the proper data/structure for the algorithm. In the course, all the time, we get a well cooked datasets...

This was a great course: challenging and insightful. I liked the way in which TensorFlow was connected to published networks/papers. I did, however, feel that there was a significant change in effort required--compared to the earlier courses in the specialization. The increased effort was mostly due to the additional reading from the literature.

The lectures were outstanding for understanding the theory but the programming assignments had almost everything pre implemented. I understand that these algorithms require a very long implementation but that just made the programming assignment as just filling the blanks and didn't make me capable of writing code for these algorithms on my own.

I would very much like to give 5 stars, but because there are mistakes in the quizzes as well was in the assignments which are not corrected, I must decrease this rating by one star. I don't find it very appealing that I must find answers for quiz bugs from the discussion board.

So please, update the material whenever mistakes and bugs are found.

Course is amazing, teaches you a lot for ConvNets, image recognition, verification, building simple models in a couple of minutes, and refining them. The only drawback is that there are errors here and there, but fortunately they are being addressed, so future learners might experience less problems. Even with this, it was a really nice course.

Absolutely fantastic course,I just loved it.....Only problem was to me in the face recognition topic.

Training a siamese network need sharing base model with multiple inputs,which is important for training model with unconventional loss functions like triplet loss.And this isn't covered in the course.If that is included in it would be awesome

Again an excellent course. Great insights in convolutional networks. The programming exercises should use ore recent TensorFlow version as the functions cannot always be found anymore in the documentation, making the tf exercises hard to make - and at a certain moment it becomes a bit trial and error instead of a result of logical thinking.

Very informative and helpful to understand the fundamentals.

The exercises could have been designed better to understand TensorFlow, while one or two more exercises for the framework could have helped improved understanding of the framework. Also, several exercises feel like being spoon-fed and do not add much to the knowledge about CNNs.

Nice Course and it covers lot of details about the current concepts in deep learning. A little more details into YOLO and NST which included how we can train them ourself instead of using a pretrained model would have been better. A little more details about tensorflow and keras implementation of the algorithms could make it more helpful.

Overall a good course in understanding some of the concepts in the world of Detection of Images, Identifying them and building a fence around them to see what are their dimensions and how to ensure we dont fail to identify them and also identify humans which is becoming more and more prevalent in more and more countries across the world.

Week 3 of the course was a bit tough, well for two reasons. Firstly I thought the exercises had long explanations and too much detail which really needed much attention to retain the key information and be able to apply it. Secondly, I have been doing this non-stop for the last couple of weeks so at this stage i think I'm also exhausted.

Extremely interesting bit a bit too high level compared to the 3 previous ones. Convolutional Networks usage are well described but the technical implementation from scratch is incomplete which makes us dependent on tools such as tensorflow or keras. An additional video showing the maths behind the complete backprop would be a real plus.

Would have liked more explicit math, maybe as optional material, for some of the "you don't need to understand the math" parts. Also, there were some errors / inconsistencies in a couple lectures. (See the forums for more info.)Overall, though, a very good course - and much "meatier" than some of the ones that preceded it in this series.

The facial recognition and verification stuff was pretty cool and I'm glad it was included. I wish the python grader was implemented correctly and there were some technical difficulties with the Jupyter notebooks. The course opened later than expected which made me lose a month's worth of fees before a notice was finally sent out. :(

This course is amazing, but it just needs to be split up into a bigger number of weeks, have much more quizzes, like every 3 videos there's a quiz, but the problem is that you have to watch about 2 hours straight until you reach that quiz, it is hard to maintain your focus abilities in these circumstances. Overall rating? Amazing!

Awesome course ! I got to understand how some awesome applications like neural style transfer really work, implement the resnet 50 which is state of the art and most important I learned how to know which feature of an image an individual neuron in a given layer actually learns ! Thank you all for the effort, you did a great job !

Liked first three weeks but not the fourth one. There are couple of reasons why I didn't like much 1) car detection totally failed when i tested with my images. Images were clear and with proper angle (as if taken from dashboard camera). 2) face recognition assignment wasted lots of time due to incorrect data and expected result!

It was a good course and Andrew did a fantastic job of explaining all the concepts in an accessible manner. I do wish he had gone more in depth about backpropagation and I also felt that the assignments towards the end were dumbed down and hence I don't really feel like I have as good a grasp on the topics as I would have liked.

I think there is an error in the assignment for neural style transfer. My code was correct and was also graded correct, but one of the expected output - total cost was wrong. Wasted a lot of time searching for the cause. On the whole the course was still great. Like Andrew's enthusiasm and lots of examples from the industry.

This course really gave feel of deep learning. The way Andrew Ng taught and content of course is preciously valuable. star less because the language Keras and Tensor Flow where I felt difficulty. You should provide language study material with examples. I hope you would implement it. It will help to work with language with ease.

The content of the lectures is great, introducing state-of-the-art solutions in a way everyone can understand. However the grader for the assignments needs fixing. Many times I had a correct solution, spent a lot of time trying to fix it only to discover via the discussion forum that an incorrect solution passes the grader ;/

A great introduction to convolutional networks. Highly recommended to learn about the cutting edge scale and depth. Minor suggestion: The Jupiter interface seems to be buggy on chrome (mac). Doing a save and check point can help relieve this problem (at least re-coding wont be needed). It would be nice if a fix is available.

Excellent videos and quizzes to learn or review CNN concepts.

The notebooks however should be refresh using modern frameworks (Tensorflow2 or Pytorch). The explanations inside are great, but there are also too guided: more freedom should be given to implement methods with only rigid formats regarding submission evaluation.