Hi everyone, I’ve started out learning Machine Learning and Deep Learning for just severial months. Lately, I’ve been super intrigued by Computer Vision and how it goes way beyond just using deep learning models.
I know that it is more than just apply Deep Learning to train model in Computer Vision. There’s a whole bunch to explore like Image Operation, Image Preprocessing, Edge Dectection, Thresholding, and more cool stuff like that.
I’m wondering, though, if focusing mainly on Deep Learning is enough for jobs or projects in Computer Vision. Should I also learn about the more traditional aspects?
By the way, any good book recommendations for someone like me who’s starting out?
Thanks a bunch.
What exactly do you mean by Computer Vision? Or, what tasks are you interested in accomplishing?
Several of the items you list are more like image processing techniques, not really part of machine learning.
I have a very keen interest in Computer Vision. If you are looking for a C.V. book, I recommend “Computer Vision: Algorithms and Applications” 2nd Edition by Richard Szeliski. This book is full of concepts, it might help you.
I’m quite new to Computer Vision, and I’m really getting into using Deep Learning for it. So, if I’ve misunderstood any technical terms, please bear with me!
Those techniques have caught my eye, and I’m wondering if I should focus on them too. Like, do I really need to learn these techniques alongside Deep Learning? Are these techniques commonly used to enhance Computer Vision projects?
It’s good to know a variety of techniques and methods. They may be handy tools for any specific tasks.
Computer Vision is a vast and very interesting subject. It involves all the techniques that computers use to interpret and understand visual information starting from Image processing, Video processing, Image/video analytics, traditional techniques for modern problems like object detection and recognition etc, and deep learning.
Earlier, before Deep Learning was popular, scientists came up with several traditional techniques to solve complex problems but with the introduction of Deep Learning, it outperformed all the traditional techniques in solving complex problems efficiently and accurately.
But we always want to do better and make our DL models better. As everyone says, making the quality of data better is the key. For example, sometimes we can derive additional features from our images using traditional techniques which can then be passed on to our DL models etc, we can enhance the accuracy. And also, for smaller/simpler problems we can simply solve them by traditional techniques rather than depending on ML. So, yes, learning traditional CV can help you to enhance your CV projects.
Regarding resources, the book suggested by @Arif_Hassan is the best CV book in the market and also you can refer to this YouTube lecture series - “First Principles of Computer Vision” from the Columbia University (Also available as a Coursera Specialization).
I wanted to shoot you a quick thank you for the awesome response. Your insights into Computer Vision and how traditional techniques can complement deep learning really helped me understand things better.