Great course for kickoff into the world of CNN's. Gives a nice overview of existing architectures and certain applications of CNN's as well as giving some solid background in how they work internally.
Great Course Overall\n\nOne thing is that some videos are not edited properly so Andrew repeats the same thing, again and again, other than that great and simple explanation of such complicated tasks.
par Paulo A F•
Great course. It has all the main state-of-the-art approaches. I just missed dealing with 3D data (RGB-D and point clouds). I believe the programming assignments get better as the course progresses because they get more demanding.
This is a great overview course. I suggest anyone interested in deep learning vision to start with this course and then move on to implement a CNN in tensor flow form scratch using one of many tutorials online.
Thank to the team for this great course!
par Matei I•
A lot of quality content in this course. The first half focuses on the intuition behind ConvNets and their implementation, while the second half focuses on applications. I thought that the neural style transfer application was particularly enjoyable. My only suggestion for improvement is to let the students do more work in the assignments for the last two weeks. I feel that most of the code in these assignments was black boxed, and I got to implement a minimal portion of the algorithms.
par Martin B•
As with all the other courses by Andrew Ng, pacing and presentation are perfect. Learning this material is highly rewarding. Programming assignments are clear and accessible, although a little bit more thorough introduction in the use of Keras and Tensorflow wouldn't hurt in some cases. I found myself pretty deep in the documentation of both libraries - although that might be part of the intended learning process. Highly recommended! - Thanks to professor Ng for making this available
par Camilo G•
Curso excelente. Da todos los detalles más importantes sobre redes convolucionales, incluyendo las matemáticas que las hacen funcionar (incluso explica backpropagation en un ejercicio opcional) y cuáles son y cómo funcionan las aplicaciones más importantes. Omite una que otra cosa, por ejemplo cómo aplicar vectorización a todos los ejemplos de entrenamiento, y de vez en cuando durante los videos secciones de audio se repiten por alguna razón, pero mayormente está bastante completo.
par Mihai L•
This course is still amazing. Finally understood what CNN's are for and how to use them.
This is the first time in deeplearning.ai specialization that I had to consult the forums. by far implementing in low level code convolutions (first asignment) was the most difficult part.
Spent more time then with the other courses but it was time well spent. Again Andrew NG delivers a good course.
The minor editing problems in videos are the only issue that might be raised with this course .
par Andrew K•
The entire course is great, from the lectures by Andrew Ng, to the homework assignments, and the TA's help on the forums. The really terrible part of the course is the coursera grader, which I had to hack for 3+ hours just to pass an assignment. I dont wanna dink the review for this because the class itself is wonderful. But please fix those technical issues. So the 5 stars come from averaging 10 stars from the course itself, and 0 star for coursera technical issues. :-)
par Omar S M•
This is an excellent course in which Professor Andrew Ng explains the concepts of convolution, pooling and convolutional neural networks very well. Also the various advanced convolutional network architectures and various applications in computer vision are discussed in an excellent manner along with references to the research papers on which the content is based. The programming assignments are also excellent and really help you learn the principal concepts and techniques.
Before taking this course, I thought computer vision had a difficult learning curve. After taking it, I found that many difficulty materials are omitted so that I could learn without too much pressure. While I could still look into algorithm details because many papers are recommended. The programming assignments cost me a little more time than the previous courses, but bring so much more fun! I felt quite proud of myself when I successfully built the CNN in my assignments.
par Ashwini J•
Thanks to Andrew Ng and team for putting together great content around Convolutional Neural Network. This is a fairly complex course, I needed to go beyond content provided in this course, specifically around understanding dimensions resulting from a convolution operation applied on an input image. This could be because it is hard to imagine a 4-d object. Otherwise, good content put together, assignments are good and useful starting point for projects in actual practice
par Shyam C N•
This course was one of the better ones in the specialization. I enjoyed it very much. The assignments are a bit more practical, and require some thought while debugging. Although some TensorFlow experience from Course 2 is expected and useful, this course requires some additional reading of the TF and Keras manuals. My only suggestion to the development team would be that they improve the NST assignment's introduction of TF methods like assign() and InteractiveSession.
par Selina N•
It's an exciting course. I find very interesting to learn object detection, facial expression and face recognition. The concept of neural style transfer is easy to understand and funny to generate image to absorb the style from another image. The explanation is useful. One improvement is some assignments only import the trained models with extra source code. It would be better for students to build by themselves to go through the whole model development step by step.
par Rahul K•
Very intricately explained course! Prof. Andrew has gone the extra mile here, making sure that the basics of CNNs have been imbibed thoroughly. Kudos to the programming assignments - They're undoubtedly the toughest of all the former deeplearning.ai courses. Use the discussion forums to help get subtle hints. I now feel that I can read CNN-related papers and even work on CNN applications. Plus, you learn how to implement Neural Style Transfer (DeepDream) here!
CNN is a tough topic to fully demonstrate. From my perspective, the lecturer simply offer an intuitive introduction and pick up some notable variant like ResNet, and illustrate the main ideas through delicately chosen case studies. That's somewhat "clever", I think. Maybe that's not appropriate, but I mean that it's friendly to a fresh learner but far from detailed and enlightening for an advanced learner. Anyway, I get to dive deeper into this field myself.
As in every class taught by him, Professor Andrew Ng makes Deep Learning concepts and applications accessible. His clear explanations during the videos lead from learning the foundations to implementing modern-architecture Convolutional Neural Networks. He provides additional information about whether certain techniques are currently utilized in research and production which bring an important relevancy to the material. Thank you for offering this course.
par CHEW L W B•
Great intro to CNNs, how they work, how to use them and the types of problems they are good at solving. I'm glad Prof Andrew Ng touched on more advanced topics such as image detection, localisation and face verification/detection and how CNNs can be applied to such use-cases. The programming problems were challenging but not overwhelming, as long as one is willing to spend some time to understand the concepts presented and explained in the lectures.
Great course! Gives a great boost in understanding of deep learning usage while solving computer vision tasks. Different ConvNet architectures, their application, state of the art algorithms are explained in detail. Sometimes there were issues while solving programming assingments, specially at the last week, but I truly appreciate deeplearning.ai work that gives everyone the ability to learn about this things very effectively. So 5 for this course.
par TANVEER M•
The course gives the basic understanding of convolutional neural network in a lucid manner.Every concept is very nicely explained. I was having some confusion with yolo algorithm which got cleared.Also Neural Style transfer and Face verification using Siamese network were the two which I haven't heard before were very interesting. The assignments are awesome where how yolo and neural style transfer works made my concepts clear to a lot of extent.
par Anshul M•
Great introduction into some of the recent and cutting edge work in the field of computer vision. The course's mathematical focus is good to understand the mechanics behind the use cases at the same time I liked the intuition about the steps in the process were shared from time to time for better context. Would have loved to get hands dirty on training models or tuning hyper-parameters - but understand it would need additional resources GPU etc.
par Amit B•
Excellent Course. It has given me an immense insight into CNN and its practical applications. I have become that much more knowledgeable thanks to this course and its contents. Sincerely appreciate the concerted efforts of the team to lucidly explain the nuances of various concepts and at the same time provide ample opportunities to the trainees on hone their skills on practical aspects of implementing the algorithms. Kudos of all stake-holders.
par Matthew J C•
Another fantastic course from Dr. Ng. In addition to object classification/recognition (which class does the object belong to?) this course should get you started with object detection (where in the picture is/are this object/s?). This course does not cover single or multiple instance semantic segmentation. Take this course (much of the coding is from scratch) & then go look at examples from your favorite API (Keras, TensorFlow, PyTorch, etc).
par Hermes R S A•
There is a dedication, from the professor and the team, to teach you the most recent developments, without skipping important introductory level concepts. Having a grasp on the Imagenet winning architectures was really rewarding. The only down side was the YOLO algorithm assignment, because the notebook was a little confusing and disorganized, but you ca get the key ideas from it. All in all, it was my favorite course on this specialization.
par Joshy J•
This is the best course for those who are serious about Deep Learning and computer vision. Some of the features of the course are Well Arranged, Simple, give a deep understanding of the mechanism, etc. We will learn Image processing, Image detection, Object detection, Face recognition and face detection through this course. Weekly assignments in the course give hand-o experience with the popular deep learning frameworks and neural networks.
par Shuai X•
Prior courses are almost all covered in the Stanford Machine Learning Course, which is free. If you don't want to waste time going through what the Stanford Machine Learning Course can offer, then this is the point to start to subscribe. Though it estimates 4 weeks of learning is needed, you can probably finish this course in a week. Assignments on CovNets and ResNets written in Tensorflow and Keras are mostly very good and very useful.
par Ashutosh P•
This is a really comprehensive course by professor Andrew Ng. He dove down to even the smallest details, you'll realize this when you listen to the lectures carefully. Make notes of each lecture as it's a long course and there are lots of terminologies in which you could easily lose yourself, stranded somewhere in between lectures having no clue what he's talking about. All-in-all, it's easily one of the best courses I've done on CNNs.
par Azer D•
Course was so helpful to understand concepts of conv nets. Also i like that Prof. Ng prepared the course with related successful papers of conv net world.One thing that i'm not happy is Coursera's Jupyter Notebook hub which I usually have problem with user authentication. Because of that I saved notebooks to my local machine, worked locally, and after completing it pasted my answers to notebook. I hope problems will be fixed soon.