Hi,
I’m trying to get my hands dirtier with Computer Vision
I really liked the ImageNet bundle offerings of this book. Do you think it’s a good idea to pursue this?
https://www.pyimagesearch.com/deep-learning-computer-vision-python-book/
Sanyam
3 Likes
I can tell you about my strategy.
I have been very much into reading everything I would find interesting and doing every MOOC imaginable in the world. There were times when I was doing 6+ MOOCs.
I think to some extent this worked, but very quickly this decays into just doing stuff, completing exercise after exercise without getting anywhere. Same goes for books- I would read books on math (linear algebra, analysis, etc), maybe learning to some extent new things but none of this translated to what I could do.
With fast.ai things are different. There are days when I am able to start working on something only after midnight or only for short periods of time during the day, but I focus on doing and not running in the fruitful treadmill of superficial learning. I still have a very long way to go to be able to say I have some valuable skills in the area of data science, and maybe I will never get there, maybe my approach is flawed again in some way I don’t understand yet, though it seems to be working for me thus far much better than what I did in the past. I learn practical things and lo and behold - I am making submissions to kaggle, wrangling with data and with my compute environment 
Didn’t foresee for example that IO would be such a big issue with larger amounts of data - spent a lot of time experimenting with it, started writing a detailed post about what I learned from Jeremy and through my own research. I wouldn’t have anticipated this being an issue and no book as far as I know tells you how to deal with this - you only encounter such problems when you prioritize doing vs ‘learning’, that is learning in this very academic and not practical sense.
I do continue to buy books (those are the only gifts I asked for Christmas) but I mostly keep them on the shelf now. Not enough mental capacity to read them at the moment. I have several posts in the works and want to work on kaggle submissions for the favorita competition, but I like to think that I will get around to reading one of the books that I have sooner than later I don’t have a lot of belongings, don’t care very much for things - some people like watches / gadgets, whatnot - so if I keep a couple even relatively pricey books on the shelf that should not be a big deal
This is what my strategy is right now but we are all in different phases of our lives and can care for different things! Not sure what the right answer is for you with regards to this book, but I am sure you will be able to figure it out yourself and if you want to ‘get your hands dirtier with Computer Vision’ I don’t think there is any place better for that in the whole entire world then fast.ai so you are in the right place
10 Likes
i used to follow Adrian - his posts are very thorough and detailed (sometimes too much ) and were very helpful when i just started with openCV about a year ago… So when he advertised a kickstarter project for this book i got all excited but the suggested TOC looked disappointing - just not deep enough tbh - i think for complete novice it might be good but you’re way beyond this point.
4 Likes
This link has a good amount of info on computer vision, worth a look.
3 Likes
@radek People like you sir, are a blessing to the community.
Thanks for sharing. I sometimes struggle with the advanced discussions and the Math of papers which everyone here is an expert in implementing. Hence I search for a ‘lower barrier’-thus I stumbled into the book.
I’m not sure now that you’ve mentioned if I should pursue it.
I think my strategy now, given that I’ve just graduated from the Udacity Deep Learning Nanodegree and have time to learn a lot would be: strengthen my Fast AI learnings now, and at the same time I hope to enroll for Self Driving Car curriculum with the end goal to find a path as a DL/CV nerd in the field (not exactly a big robotics fan) and gear up for the Part 2 with a hope to improve upon my Coding skills at the same time.
Definitely sticking to more doing in 2018 rather than just learning.
2 Likes
Thanks! I think I’m not going to go for it now. Would try to replicate the table of contents as an year’s end goal by discussing in a few communities and let’s see how it’d go.
I feel as a noob to the field and an undergradute-there’s this fear of missing out on learning that- Oh he teaches creating an API, hmm maybe I should go for it.
Oh this book is on creating Object Search Engines-Maybe I should go for it too, I’m certainly not ready to implement papers and create a ‘model from scratch’ but I certainly hope to do the same. While at times I struggle with programming (re referring Numpy, PyTorch documentations) this goal seems intimidating every now and then. Hence ‘find a lower barrier’ approach.
1 Like
in practical terms it seems that pretty much all CV related jobs would require hands on C++ - which is a beast within its own right and in a sense orthogonal to the DL stuff
oh and i assume you’re aware of the MIT SDC course that starts shortly, in Jan
1 Like
Oh. I was under the impression that Python is being equally explored, I have seniors working on open cv with Python. And the startups I’ve spoken to have told me they use Python+OpenCV Pipelines.
Plus, I hope to see shift to Python given that open cv is growing and has a c++ implementation and a Python api.
And yes, thank you for pointing that out. They’ve a slack group and they will start putting out content by next week. I think I’ll be following both the worlds-Udacity and MIT’s OCW to get the best out of them.
Edit: If anyone is interested, Here is the Slack invite
1 Like
ok, sure… i just did some quick check on github and indeed for CV it 50/50 bw C++ and Python but say for SLAM it’s more C++
1 Like
I believe for SLAM, most of the use-cases are obtained using ROS.
Surprisingly, Udacity and MIT both mention using Python during the course for the Vision modules.(MIT-for the complete course, Udacity will use C++ for Path Planning)