Everything else is perfect, except the W3 assignment submission took me 3 extra hours to be correctly graded (my results match the expected results, and I didn't violate any rule that FAQ mentioned, but grader just gives 0. I read through the forum threads, tried all approaches, and finally get it passed after having Coursera helpdesk load a fresh notebook for me and type in all over again, a little bit frustrating).

Very interesting and well structured course. Great lectures and content.

Only critics: The actual tasks to be computed/coded in the assignments did not include the "interesting"-ML issues. A lot of linear algebra coding was asked for (i.e. compute loss-function several-times), but which did not help me to understand the underlying ML / CNN principles. A little bit more effort could be done there!

Thx for the course!

Very interesting course , not as the same level of the first two courses BUT it is an excellent resource to get in the Deep learning world. I did not like that they don't teach you how to use tensorflow and keras in a more concise way. And sometimes the content doesn't seem as usefull as the first two courses and more in the deep learning world where computer vision techniques become obsolete so fast...

solid 3.75/5

This course was perfect! I finally understood convolutional neural networks and its popular architectures. Implementing CNNs in numpy was a useful exercise as well. Andrew once again proves to be the best teacher of ML. The reason for four and not five stars is the number of technical glitches throughout the course especially in the final programming assignment. I'm sure they can be improved upon for future classes.

The material is meaningful and instructive.

I did well in a capstone project of image recognition in school with the knowledge from the whole specialization (especially skills on Conv-net in this course).

However, it would be better if we could have access to more computational power and really play with large data set and complicated algorithms instead of just doing simple maths and loading models to see the result.

This course is useful for getting acquainted with the terminology and main ideas that are used in building and training convolutional neural networks, as well as learning the names of a few most well-known existing networks. However, I didn't find the practical exercises to be very useful or relevant -- at least, in comparison with exercises in first two DeepLearning.ai's courses in the same specialization.

A very complete and diverse course with several applications of deep networks. Only problem was the assignments which had a lot of problems in the grader, and created a lot of frustration and lost time figuring out where the problems were. Otherwise it would have undoubtedly been a 5 star course :-)

Keep up the amazing work you're doing, especially for Course 5 where I have great expectations!

Cheers

António

Great course to understand the fundamentals of CNNs and various CNN architectures currently used in the field. Would have liked to get a little better implementation wise by doing this course in terms of some architectures, but probably it will take a lot of time and is infeasible in the course. All in all, this is a great course to learn some of the best techniques in Deep Learning for Computer Vision.

The course material is very good by itself, but the issues with programming assignments spoiled the overall impression. Discussion forums are full of messages about people struggling with the grader or with functions provided by the course and no assistance from teaching staff. You should pay more attention to supporting one of the top courses on Coursera.

Aside from that - Andrew was great as usual :)

There is a feeling of doing some dark magic on every exercise around images. Maybe putting more emphasis on how to load/process images would be great for people like me that don't know about this world. Also, the part of using a pre-trained model could be improved as well: maybe pointing students about how to search/download it from the right place, and providing more detail about what you download.

The content is interesting, practical and relevant to anyone interested in how CNNs are used in computer vision. Unfortunately, the course materials still have a couple of bugs such as videos that aren't edited well and buggy programming assignments. Luckily, the Discussion Forums offer hints on how to resolve these issues, but I'd expect the course creators to address these ongoing issues faster.

Good course but some parts lack clarity and require too much second guessing. This is particularly true for week 1 assignments, where we are supposed to move from the TensorFlow Sequential to the more flexible API (the way it describes how to add layers to a model is short of clarity about adding arguments at the end, as typically you observe layers successively applied to (x) in most examples).

I think the course videos and lectures are awesome and very informative. But the quizzes and programming exercises are very simple and not involving enough. I think the instructors have to make it a little more rigorous to challenge students as they prepare for the real-world tasks. In my opinion the hand-holding is great but the instructors should have students code for some simple dataset.

It's a really great course covering important concepts in CNN such as residual network, face recognition, neural style transfer and other very captivating topics. The only complaints I have about this course is that the programming assignments are a little too simple, most of it is already done and we are only required to do a very small part. I would have enjoyed more challenging homework.

Course is good. Much thanks for Angrew Ng, to explaining CNN in simpler way. However, the practical assignments are not properly configured to load. In Week2 & Week4 the practical assignments did not load properly. Hence it took longer time for me.

Also, couple functions' expected output and my implementation output did not match however, I got the full score. This again misleads us.

The lectures are what you would expect from Andrew Ng. Excellent.

Some of the assignments make unreasonable jumps in expectations regarding understanding of TensorFlow and Keras operations. An overview of exactly what is being used would be helpful as some of it is very nonintuitive.

Additionally, some of the assignments have known grading issues. Be sure to check the discussions.

Overall the course is solid and covers many important topics ranging in complexity from simple to advanced and state of the art (NST). However, Videos are sometimes not properly edited. There is repetition of dialogue. Also, I think practice questions can be made a bit more challenging. I also noticed that in this course there aren't any explanations for right or wrong answers.

Excellent lecture content.

Some of the programming assignments are quite poor. Sometimes there are minor mistakes in function descriptions, and other times the whole assignment architecture/plan is not well thought out. If the staff doesn't have resources to improve this, then allow the community to create branches and submit merge requests :)

Overall, I'm happy with this course.

The Course is more than great, learning about using ConvNets in different problems and applications was very interesting and useful. The only drawback is that the code in assignments is built on TensorFlow 1.x which is outdated and even some links to TensorFlow or Keras documentations are note working, I'm sure this code will get upgraded soon. Thanks Andrew and all the staff.

Andrew NG explains CNN fundamentals really well in this course. I liked the use-case based teaching. Also, the assignments were at par with the lectures. I faced a couple of issues in Face recognition assignment of Week 4. The team should look into that. Looking at Discussion forums helped me in moving past those issues. A big shout out to people actively participating there.

Learned a lot. Theoretically, course is must-to-do, most of the code is done so please do not expect that you'll be a king in Tensorflow and CNN. However, you will be able to implement things in real-time and yes coding you can learn anytime at your own pace. Also, the projects that were implemented in the exercises are not something that can be done alone (at least for me).

Initial introduction of convolutional neural networks was very thorough, with week one even addressing back propagation along convolutional neural networks via the programming exercise. Later weeks showed interesting ways in which the theory of convolutional neural networks has been applied, although some independent research has to be to supplement learning in the course.

Downsides are some edition errors in videos (like not taking out parts where Andrew repeats same phrases) and erros on quizzes' gradings (marking as wrong a correctly answered question and giving something like "answered out of alloted time" when everything was submitted within the time limit). But other than that, this is a great introductory course to computer vision.

Good course with very relevant and practical content. Since it was the first time this course was offered, a few bugs in the assignments notebooks. While you don't need to be a guru in vector algebra to complete this course, I would appreciate a little more focus on the rationale for using 'Axis = True', for example or 'Keep Dims = ???. thanks for a great course.