Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Ask HN: What are the best resources to learn computer vision?

news.ycombinator.com

Post reply on HN
Print view View original

41–50 of 59 posts

Re: Ask HN: What are the best resources to learn computer vision?

#42

Post by thinkMOAR »

lightbyte wrote: post #37

Surprised nobody has posted http://course.fast.ai/ yet. I've been following along with it so far for the first 4 lessons and it has been extremely helpful in understanding how deep learning works from the perspective of someone who did not have much of any related baseline knowledge except how to program. Jeremy is an excellent practical teacher.

Happy it is interesting for you.

I too got this url referred by somebody, and I got excited after their extended intro why, how etc their course different and better then any other.

Though after 5 videos i know nothing more then from any other ML/AI guide on the internet then i did before. 99% is only related to image classifying, and i'm simply seeing too many guides for that.

If anybody has some good links/videos on ML/AI on structured data, please comment and i'll be thankful and happy to click 'm :)

Re: Ask HN: What are the best resources to learn computer vision?

#43

Post by visarga »

You don't specialize in surgery before learning biology. Similarly, you don't specialize in CV before learning basic ML and DL. The fundamental concepts are the same no matter if the modality is text, image or video (for example: regularization, loss, cross validation, bias, variance, activation functions, KL divergence, embeddings, sparsity - all are non-trivial concepts that can't be grasped in a few minutes, and are not specific to CV alone).

Re: Ask HN: What are the best resources to learn computer vision?

#44

Post by zionsrogue »

zelon88 wrote: post #41

PyImageSearch by Adrian Rosebrock. http://www.pyimagesearch.com/

Adrian here, author of PyImageSearch. Thanks for mentioning the blog. If anyone has any questions regarding learning computer vision, please see my reply to "sphix0r" below.

Re: Ask HN: What are the best resources to learn computer vision?

#45

Post by thearn4 »

a_d wrote: post #21

Is OpenCV (traditional CV technique) better to use or Deep Learning based approach? Has anyone done a comparison of the two approaches? The obvious flaw with deep learning is that it requires large labeled data sets - but assuming that is available, which one is more accurate at object detection (hotdog or not), detecting features on an image (faces, manufacturing defects)?

I use OpenCV for reading and writing real-time video streams (like webcams or video frames) for my hobby computer vision projects, but tend to use other ML or image-specific libraries for actual processing. The cascade classifiers in OpenCV are okay, but it isn't too difficult to set up something comparable in scikit-learn that is more modern and robust. Though a bit less performant if you do need real-time response though.

Best example that I have is a pulse rate detector that I put together, that uses OpenCV for video frame extraction & display but bare numpy/scipy for the rest.

https://github.com/thearn/webcam-pulse-detector

Re: Ask HN: What are the best resources to learn computer vision?

#49

Post by wyc »

This is pretty old school, but I recommend Multiple View Geometry by Hartley and Zisserman (http://www.robots.ox.ac.uk/~vgg/hzbook/) to get through the fundamentals...it's really good to understand the geometric foundations for the past 4 decades. Along the same lines, you have Introductory Techniques for 3-D Computer Vision by Trucco and Verri (https://www.amazon.com/Introductory-Techniques-3-D-Computer-...), which also goes over the geometry and the fundamental problems that computer vision algorithms try to solve. It often does come down to just applying simple geometry; getting good enough data to run that model is challenging.

If you just throw everything into a neural network, then you won't really understand the breadth of the problems you're solving, and you'll be therefore ignorant of the limitations of your hammer. While NNs are incredibly useful, I think a deep understanding of the core problems is essential to know how to use NNs effectively in a particular domain.

After getting a grip on those concepts, Szeliski's Computer Vision: Algorithms and Applications (http://szeliski.org/Book/) had some really amazing coverage of CV in practice. Mastering OpenCV (https://www.amazon.com/Mastering-OpenCV-Daniel-Lelis-Baggio/...) was very useful when actually implementing some algorithms.

Re: Ask HN: What are the best resources to learn computer vision?

#50

Post by mendeza »

Here is a guide I have developed over 6 years when I dove into copmuter vision around 2011. My path has been self taught until recently I took a graduate course.

I started from wanting to develop AR apps during my undergrad, Here are the best resources I have found to date:

Computer Vision is very theoretical and experimental, so the more hands on, the better! My approach has been to go top-down, overview the landscape and slowly progress deeper.

Begin with the best library for CV in my opinion: OpenCV. The tutorials are amazing!

Python tutorials: http://docs.opencv.org/3.0-beta/doc/py_tutorials/py_tutorial...

C++ tutorials: http://docs.opencv.org/3.0-beta/doc/tutorials/tutorials.html

Immerse yourself in these and build any apps you think of!

Then go into: pyimagesearch tutorials http://www.pyimagesearch.com/ and aishack.in http://aishack.in/,

tons of great tutorials to learn different topics of vision with coding walkthroughs. Understand the examples and rewrite applications.

Then Dive Deep:

Get the new OpenCV3 book, a nice deep overview of many topics in computer vision. https://www.amazon.com/Learning-OpenCV-Computer-Vision-Libra...

And watch this course on youtube:

https://www.youtube.com/watch?v=skaQfPQFSyY&list=PL4B3F8D4A5...

I feel like then, you will have so much exposure that when you dive into formal classes and textbooks, you will really understand and be enlightened.

This was the general way I learned computer vision, and recently I completed a cv internship for nanit.com . I was not hired for my formal knowledge, but they were impressed by all the various projects ive done and knowledge I had on many vision topics.

I also recently took a formal course of vision at Cornell: http://www.cs.cornell.edu/courses/cs5670/2017sp/

All the assignments have starter code in python and opencv. This was an amazing class as it dove deep into 3D computer vision, which is so relevant to augmented reality!

Also, here is a link opencv examples for iOS: https://github.com/Itseez/opencv_for_ios_book_samples

here are links for opencv example for Android: https://web.stanford.edu/class/ee368/Android/

Hope this helps! Shoot me a dm if you or anyone has more questions!

Post reply on HN