I don't understand the purpose of listing a handful of research papers and not teaching the topics

I completed all three courses for the GAN specialization. Overall, this is an excellent course. The content is high quality and compact. The course is highly recommended for professionals who have limited time to keep up with the state-of-the-art in GANs. I feel that the course has given me enough knowledge for me to find ways to apply these skills for good in the industry.

Areas for possible improvement: 1. Some of the lab exercises put focus on the wrong areas. In some cases, I feel like I was spending time on tensor manipulation instead of learning the important nuances of the algorithms. 2. I would love to see the course extended. It's relatively short and I think some of the advanced optional content could be incorporated into the standard curriculum. What I value most from this course is how it condenses and simplifies concepts. The optional content leaves the reader to self study and doesn't help with accelerating learning. Insights that help the learner understand the architecture differences, improvements as well as the pros/cons of the GANs referenced in the optional content would be valuable.

The specialization contains excellent theory, but is extremely lackluster in the assignments department. If you are fluent with PyTorch - you will be fine. If, like me, you're only familiar with Tensorflow or other ML libraries - it might be a struggle.

The course itself provides next to none code explanations. A lot of practical assignments end up becoming excercises in reverse-engineering their testing code. Reading through all the questions on Slack, I am far from alone in this. Some code cells give you tasks along the lines of "you gotta do this, there are a lot of ways to do it, so do it somehow".

Bottom line. Was the course useful? Yeah, I will implement things I learned here in my GANs. Was it a pleasant learning experience? No, it was frustrating due to a glaring lack of code explanations.

All it would take to make it much, much better - have an extra video per week which would go over putting the new theory into code, like many other courses here do.

Nice explanations. All you need to know about the state of the art in GANs.

Why do you need to start a course by insulting your students with some "oath"? You don't own the knowledge: there are github repositories and papers available online. All we need is a good introduction to the topic. Which you did provide, by the way, perhaps not as detailed as I wanted, but there was interesting material.

Instructor is very clear in teaching. It is too precise without sufficient fundamentals. Have been struggling for the program assignment. The program itself is good example, but the part for fill in is not well designed, and often stacked in something that is NOT related to the GAN model technique but data structure use or pytorch use and spent huge amount of time to figure it out... code downloaded of pdf and ipynb don't work though you may figure out to covert json file to ipynb

If you have better options skip this, it will save your time and money.

Rushes through complicated topics without really explaining them

01. Very well crafted course content. 02. Well delivered lectures. 03. Very good division of compulsory and optional course material. Comment: In a specialization we cover a lot of stuff. Many things that we learn early on get superseded by more advanced material during the course and otherwise also towards the end, information gets mixed up. It may be a good idea to include a concluding lecture as part of specialization to just recap the material covered in the course of specialization. This will be an icing on the cake.

Just completed all 3 courses. Overall it's fun to learn and play with GANs. The labs are surprisingly well designed and make it easy to get started. Even with prior knowledge in this area, I still find it valuable and informative to catch up with recent research progress, many of the cited works are published within a year. Great learning experience.

It is just great hearing the subject from a PhD owner . This course is just the right length and right difficulty for anyone who really wants to broadly "understand" the already broad subject for his/her job or research goals.

I really liked the exposure to preparing various loss functions in paired and non-paired GANs, introduction to other applications, and many great changes to improve the quality of the networks!

GANs are awesome, solving many real-world problems. Especially unsupervised things are cool. Instructors are great and to the point regarding theoretical and practical aspects. Thankyou!

Understanding of GANs: You've gained a deep understanding of the fundamental components and applications of GANs. This includes knowing how GANs work, their architecture, and the roles of the generator and discriminator networks 1 2 5 8. Building GANs: You've learned how to build and implement multiple GAN architectures using PyTorch. This includes creating basic GANs, advanced Deep Convolutional GANs (DCGANs), and conditional GANs capable of generating examples from determined categories 1 2 6 8. Training GANs: You've learned how to train GANs, including how to deal with common challenges like imbalances between the generator and discriminator, unstable training, and mode collapse. You've also learned how to apply loss functions, such as the W-Loss function, to solve the vanishing gradient problem 1 2 5 8. Evaluating GANs: You've learned how to evaluate GANs using methods like the Fréchet Inception Distance (FID) to assess the fidelity and diversity of GANs. You've also learned how to identify and detect bias in GANs 1 2 5 8. Working with Different GAN Models: You've gained experience with a variety of advanced GANs and learned how to use them to create images. You've also learned how to implement techniques associated with state-of-the-art GANs, like StyleGANs 1 2 5 8. Applying GANs to Real-World Problems: You've learned how to apply GANs to solve problems in areas like computer vision, multimedia, 3D models, and natural language processing. You've also learned how to use GANs for data augmentation and privacy preservation 1 2 5 8. Practical Experience: Through hands-on assignments, you've gained practical experience in implementing and training GANs. This includes creating a GAN model that can generate hand-written images of digits 9

I've just completed the specialization and my thoughts are that everyone should take it (that are interested in GANs! I feel Sharon is a great teacher and the entire team did a really good job on putting togethor these courses. After completing it I definitely have a much better view of GANs, their architectures, successes and limitations, and have a solid background to tackle reading papers and implementing them on my own. Thank you for making this specialization!

With all the positives (which is why I rate it 5/5) there are in my opinion things that can be improved. Especially I think there is too much hand holding for the labs, out of 100 rows of codes I code maybe 2-3%. Many of these don't give much value coding but I want to feel like I did it! Unfortunately now I am left guessing if I have truly mastered the material (and I'm quite sure I haven't, so I will need to re-implement these on my own). Also since you state that calculus and linear algebra are prerequisites then stick with it! You are trying to be too inclusive and there are several part of the courses where I thought it was entirely unecessary because everyone taken Calc and Linalg already has this knowledge. I would prefer instead if you spend this time making other videos where you go in more depth, perhaps going through some of the difficult math etc. Hopefully you try to improve this for future courses done by deeplearning.ai

This was the most challenging of the series so far. It was really great at not hand-holding as much in the programming exercises you that you get a better learning experience of actually struggling through creating your loss functions and compiling your neural network. If I could add one improvement, it would be to include some sort of capstone project wherein we would be required to implement one of the GAN architectures taught (DCGAN, StyleGAN, PatchGAN, or CycleGAN) in our own dataset or perhaps a different dataset. This would be quite challenging as the code would not be provided in terms of how to compile the network and training loops needed. This may also serve as a final challenge to figure out if we have really conceptually absorbed the different architectures and their respective limitations/implementations.

Thank you Deeplearning.ai and Coursera for offeringg this excellent specialization. I totally enjoyed the courses and can say I have been given an overview of GAN. However, the optional units were not given enough supporting explanation or time to allow the uninitiated to explore in depth. I would really like to see a followed on or alternative (Honors?) track to digest them.

There are many advanced notebooks in this course. Although it was crafted well with detailed explaination, the concepts are still relatively difficult to understand. It would be more beneficial to the students if Sharon could explain the concepts as well. Please consider a GAN course part 2 to explain the technical details. I would be very happy to pay for the course.

Excellent course videos, programming assignments, readings and optional colab notebook. Entire GAN specialization is really good to learn, understand different types of GAN architectures, losses etc. Special thanks to instructors, specialization team, deeplearning.ai and coursera platform for making this specialization available for learners.

This course is exactly what I wish a course should be: its very well structured, the assignments are evaluated and are also very well designed, and the content is really up to date with the state of art. Just fantastic course and the instructor (Sharon Zhou) pace of lectures is also really good (not too slow and not too fast). Thanks.