Over the years, I have pretty much done (and earned certificates) on all the courses Professor Andrew Ng has taught on-line (which ML enthusiast wouldn’t :-)), but while trying to understand transpose convolution, I came across the following YouTube video from Professor Ng explaining what is transpose convolution.
He also refered to a programming assignment on unet.
Right! As Deepti says, this was covered in DLS C4 (ConvNets) Week 3. There are two assignments that week and it’s the Image Segmentation with U-Net assignment that uses transpose convolutions, if you want to see them in action.
One other specialization that uses them is GANs, which is taught by Prof Sharon Zhou, not Prof Ng. It’s very well done and really interesting. One other beneficial thing about the GANs Specialization is that it uses PyTorch instead of TensorFlow (which is used everywhere else here). So in addition to the interesting topic, you also get a nice introduction to PyTorch as a side benefit.
I went to the course website, and looked at Course 4, week3, but I could not find this video. Please see the enclosed 2 graphics. Which video covers this topic?
And furthermore @paulinpaloalto, the only programming assignment I recall doing (and which is what I see even now) is “Car Detection with YOLO”
@paulinpaloalto, I don’t know whats the issue, but please see my screen shot below. I only see quiz and programming assignment after (Optional) Region Proposals video.
I recall completing the DLS soon after it was first released by Prof Ng. Could it be that in subsequent releases of DLS, some modules were updated, but on Coursera website, I get to see only the version I took? Just a thought.
You must be in an old version of the course. I don’t know why that would be. You should contact the Coursera Help Center to get moved to the current version of the course. But this change happened in April of 2021, so you are in a really old version of the course. I have no idea how that could happen.
Thx @paulinpaloalto. I did the course immediately after it was announced. I can’t recall when that was. I will check with Coursera help center. Looks like I missed a few important topics. Cheers Ram
The main purpose of the updates in April 2021 was to convert from TensorFlow 1.x to TensorFlow 2.x with Eager Execution mode. So the courses using TF (C2, C4 and C5) were updated in a major way. I don’t remember the other details, but I’m pretty sure that C1 was also updated at the same time. C3 does not involve any programming assignments, so I don’t know if it was updated.
I am tagging L.T of the deep learning specialisation for further assistance as he can also reach from his side to Coursera and provide you information if any.
As long as I remember based on your scenario, you might have to repurchase the course to get the updated version of Deep learning specialisation. If the changes/update was made within the 6-month of course subscription, you would have got access to the update version, but as you state you have taken course long back, you would need to again purchase or pay for the course, or you can try financial aid option where if coursera approves your financial aid, then some percentage of the money you are suppose to pay, and you have access to the course.
@Mubsi kindly let the learner know how the update version of DLS could be accessed by him as he has taken course long back and want to access some of the updated course material.
Hi Mr Paul, a kinda related question, Does this platform offer a CNN course thats mainly in PyTorch? Im researching into CV and most of the papers are in PyTorch so I would like to extend my knowledge there.
Thanks
As @Deepti_Prasad mentioned, you’d need to resubscribe to see all of the updated changes.
As Paul mentioned, TF version was updated, as well as 2-3 new assignments were added, new lecture videos, a whole new week on Transformers was added in C5 are some of the major changes I remember. No changes were made to C3.
The major changes happened starting from C2W3 and onwards (excluding C3).
If you have time to check for the difference between the old and the new versions, and since the major changes are in C2W3, C4 and C5, you might compare the tables of content of C2, C4, C5 (the links bring you to the ToC) with the tables of content in your current version.
I have not seen a old VS. new thread here so I can’t share that with you, but I guess it wouldn’t take you more than 10 minutes to do my above comparisons.
The only course/specialization here from DeepLearning.AI that uses PyTorch that I know is the GANs Specialization. It’s a very interesting series and very well done, taught by Prof Sharon Zhou. It’s not strictly speaking a “ConvNet” course, but GANs do use a lot of convolutions and transpose convolutions in their implementation. Definitely worth a look to expand your horizons for generative images and as a way to get a nice thorough intro to using PyTorch.
Of course Coursera is a big place and has material from lots of other providers besides DeepLearning.AI. You could try searching the Coursera catalog.
Thank you for response Paul, in the DLAI ConvNet module I saw that they have the YOLO algorithm, is it also implemented in TF? knowing the original algorithm was using Darknet and the current iteration uses PyTorch?
The YOLO assignment in DLS C4 W3 uses TensorFlow. I have no idea about Darknet or what the recent versions of YOLO use as a platform. This version of DLS C4 was published in April 2021 and the original was from October 2017.
Darknet was what Joseph Redmon called the neural net framework his team developed in C and CUDA. To the best of my knowledge there is no API and one doesn’t develop in Darknet. You can use the code and libraries or call it from scripts.
