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)?
Ask HN: What are the best resources to learn computer vision?
21–30 of 59 posts
Re: Ask HN: What are the best resources to learn computer vision?
#22Is 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)?
For the widely used handwriting dataset, MNIST, there is a site that actively tracks the best algorithms (although I'm not sure when it was last updated): http://rodrigob.github.io/are_we_there_yet/build/classificat...
You'll notice that all of the top contenders use Neural Networks, but I would bet that many of them use at least some traditional CV techniques to transform the images at various steps. That said, many of the more modern deep learning approaches are ditching CV altogether, just feeding in raw pixels without any normalization or transformation, leaving fewer parameters to tweak.
Re: Ask HN: What are the best resources to learn computer vision?
#23For a full course, Nothing beats CS231n: Convolutional Neural Networks for Visual Recognition http://cs231n.stanford.edu/ by Andrej Karpathy et al. Also, for a general and high level introduction to neural networks, I wrote a Learning Deep Learning in Keras http://p.migdal.pl/2017/04/30/teaching-deep-learning.html , focusing on visual tasks.
I did this course, but couldn't finish all the assignments. Loved it. Please note that this is a convolutional neural networks course, not computer vision as such. From what I know computer vision encompasses a variety of non machine learning based algorithms, which are not covered in this course.
As an engineer, it's difficult to know when to use deep learning and when to use more classical algorithms. Often, you have to try both and see which is better (twice the work, hooray!). The classical algorithms often are very understandable, and you can reason with what's going on and figure out what is breaking. Deep learning is so much harder, e.g., are my hyper parameters bad, or do I need another 30 GPUs running for a week?
Re: Ask HN: What are the best resources to learn computer vision?
#24Earlier quoted context omitted.
I did this course, but couldn't finish all the assignments. Loved it. Please note that this is a convolutional neural networks course, not computer vision as such. From what I know computer vision encompasses a variety of non machine learning based algorithms, which are not covered in this course.
As an engineer, it's difficult to know when to use deep learning and when to use more classical algorithms. Often, you have to try both and see which is better (twice the work, hooray!). The classical algorithms often are very understandable, and you can reason with what's going on and figure out what is breaking. Deep learning is so much harder, e.g., are my hyper parameters bad, or do I need another 30 GPUs running…
Imho, deep learning has little to do with engineering, and more with guessing, hoping and praying. But it seems you can often get something to work if you do those three things hard enough.
Re: Ask HN: What are the best resources to learn computer vision?
#25OpenCV and http://www.pyimagesearch.com/ disclaimer: not related to any of these
Adrian here, author of the PyImageSearch blog. Thank you for mentioning it, I appreciate it. If anyone has any questions about computer vision, deep learning, or OpenCV, please let me know.
In regards to OPs original question, I'm actually working on solving your very problem right now. About 1.5 years ago I created the PyImageSearch Gurus course (https://www.pyimagesearch.com/pyimagesearch-gurus/) with the aim of bridging academia with actual real-world computer vision problems. The course has helped readers in their academic careers, such as securing grants (http://www.pyimagesearch.com/2016/03/14/pyimagesearch-gurus-...) as well as students become practitioners and land jobs in the CV startup space (http://www.pyimagesearch.com/2017/06/12/pyimagesearch-gurus-...)
Within the next month I'll be launching PyImageJobs which will connect PyImageSearch readers (especially the Gurus course graduates) with companies/startups that are looking to hire.
Finally, I'm also working on my upcoming "Deep Learning for Computer Vision with Python" book (https://www.pyimagesearch.com/deep-learning-computer-vision-...) which is now 100% outlined and I'm on to the writing phase.
Definitely take a look and if you have any questions, please let me know or use the contact form on my website if you want to talk in private.
Re: Ask HN: What are the best resources to learn computer vision?
#26I found Computer Vision: Algorithms and Applications really good. You can download it for free (for personal use) at http://szeliski.org/Book/ .
+1
This is the most comprehensive book I know of on Computer Vision. The diagrams in the book (including captions) themselves do a great job of explaining things.
Re: Ask HN: What are the best resources to learn computer vision?
#27Tensorflow tutorial
Re: Ask HN: What are the best resources to learn computer vision?
#28OpenCv is a great start, and is where I began.
this book: http://shop.oreilly.com/product/9780596516130.do has a number of worked examples that explain things well.
It does touch on Machine Learning, but it focuses much more on the fundamentals of computer vision, like feature detection, that allows things like SLAM to exist.
Re: Ask HN: What are the best resources to learn computer vision?
#29I learned mostly from MATLAB documentation. Good if you want theory and implementation, and most capabilities can be done with open source equivalents if you don't have MATLAB.
Re: Ask HN: What are the best resources to learn computer vision?
#30This one is pretty good:
https://www.udemy.com/computer-vision-with-python/