Some of the best and most updated resources about neural nets for visual recognition come from the highly popular courses at Stanford. One of these is CS231N - Convolutional Neural Networks for Visual Recognition.
As part of the Massive Open Online Courses initiative, Stanford has made the entire course curricula available for anyone wanting to learn. Why would they do it?
It's not like they have anything to gain from keeping their materials closed and secret. Plus, making these materials open and free, they allow the spread of information. Does this mean that anyone going through the materials will receive the same education they would get at Stanford but without paying tuition?
Not really, going the conventional way, and taking the course at Stanford means benefiting from in-person education, which may be more efficient to many students. You'd also benefit from being in an environment and spending time and sharing thoughts about the learning resources. So, in my opinion it's still worth taking the conventional route if you have the financial means necessary.
However, having free access to the lectures is an invaluable opportunity. But, you'd have to motivate yourself going through when the going gets tough. Statistics say that the majority of learners who start a free online course, never actually finish it. So, the odds are against you!
Nevertheless, let's see what's inside this Stanford course on Visual Recognition with Neural Networks:
- principles of image classification
- optimization and the loss function
- neural networks
- convolutional neural networks and how to train them
- software used for deep learning
- recurrent neural networks
- and more.
There are 16 video lectures in this course, most of the being more than one hour long. The complete syllabus is at its Stanford page where you can also find the complete syllabus and the assignments if you want to try doing the yourself.
The course has some heavy prerequisites, so you must be proficient in Python, C/C++, you need to know linear algebra, calculus, statistics and probability theory, as well as concepts of machine learning (completing CS229 would be a +).
Even though the statistics are sobering for free online courses, in my experience, completion rates are much higher if you have to pay for it.
For example, I completed the 4 courses in the Deep Learning specialization at Coursera in about a month and half. I had to pay $50 per month and I could learn at my own pace. If I were not paying for it (audit only), I would still have access to the lectures, but not the assignments or the certificate of completion. It would've probably taken me much longer only to audit 1 course of the 4, let alone go through all of them in less than 2 months.
Also, if the course topic is of high interest to you and you are well motivated you could still go through a free course - so forget about the stats. If you want to go through CS231n, you can find the lectures on Youtube, starting with the first one below:
To stay in touch with me, follow 
Cristi Vlad Self-Experimenter and Author