Latest YouTube Video

Wednesday, January 10, 2018

Why I started a computer vision and deep learning conference

The vast majority of blog posts here on PyImageSearch are very hands-on and follow a particular pattern:

  1. We explore a problem.
  2. We write some code to solve the problem.
  3. We look at the results, explaining what worked went well, what didn’t, and how we can improve on the solution.

I love this approach as it enables me to share algorithms and techniques you can apply to your own projects and research.

However, every now and then I write a post that is much more personal and pulls back the curtains to reveal what it’s like running PyImageSearch.

Today is one of those posts.

Inside this post I’ll be giving an intimate look at PyImageConf 2018, including my thoughts and rational behind:

  • Why the first machine learning conference I attended was a total disaster…
  • …and why it was completely my fault
  • Why I started PyImageConf
  • The two “must have” characteristics of any conference I attend
  • Why you should attend PyImageConf and the value you’ll get from it

To take a behind the scenes look at PyImageSearch, along with my thought process in starting a conference, just keep reading.

The first machine learning conference I ever attended was a complete disaster

Figure 1: ICMLA 2013, the first conference I presented at, was a complete disaster. And it was totally my fault (but not for the reasons you might think).

What happened?

Basically nothing.

And it was my fault.

The first computer vision/machine learning conference I attended was the IEEE International Conference on Machine Learning and Applications (ICMLA) 2013.

The conference was held in Miami, FL from December 4th-7th 2013. It was a particularly brutal winter in Maryland that year and I was thankful for the chance to escape to somewhere warm (and enjoy the world-famous Miami beaches).

I arrived at the conference a day early to register and continue to prepare/practice my talk. I was a bit nervous to say the least as this was the first paper I ever presented. This paper would later become the crux of my PhD dissertation.

The next day I presented — I must have been the the second or third person to speak that day which further rattled my nerves.

When my time slot came around I pulled myself together and delivered the talk.

The talk went well. The paper was well received. Criticisms were minor but justified. Its was an experience pretty much every young grad student presenting their first paper could hope for.

But that’s where the positives ended.

The rest of the conference was a catastrophe and I have no one to blame but myself.

Back then I was a shy, introverted person — I struggled to approach people and have a basic conversation. And at the evening events I wasn’t comfortable using alcohol to help me loosen up and network with people. As an introvert I tended to gain my energy from alone time, that’s where I was most comfortable.

That entire conference I spoke with a total of five people. One of them was the person running the front desk. Another was a cleaning lady when I asked for more towels. Only three of them were actual conference attendees.

I spent the rest of the conference either:

  • Walking on the beach, alone
  • Sitting at a bar, alone
  • Or hauled up in my room, reading papers (again, you guessed it: alone)

2013 was a really hard year for me. There were a lot of challenging personal/family issues going on, some of which I discussed in this post. I had a lot of personal matters that prevented me from my making the most out of the conference.

But more to the point, I hadn’t learned a valuable skill:

To put yourself out there and really learn from others.

I was comfortable learning from my teachers back at the university…but not from “strangers” I just met at a conference.

On the flight home I remember being frustrated and discouraged with the experience.

And I vowed to never let it happen again.

The next conference wasn’t a disaster…I wouldn’t let it be

Figure 2: At my next conference, I vowed to never let my ICMLA experience happen again.

I got my redemption in April 2014.

This time I wasn’t at a computer vision or machine learning conference — instead I was attending a small entrepreneur conference (MicroConf) in Las Vegas.

I was on the fence about going.

At the time, my introverted self was practically making up excuses to avoid going to a highly stimulating, overwhelming venue such as Las Vegas.

But I remembered my vow from ICMLA:

I wasn’t going to let another conference be a disaster.

I packed my bags and took off for Vegas.

During the registration I chatted with Xander Castro, the conference coordinator (who is now the conference coordinator for PyImageConf, and also one of the best human beings you’ll ever meet), asking him what his favorite places to eat at in Vegas were.

I then walked up to the first group of people I saw and introduced myself, asking what each of them what they did. I spoke with the group for ~45 minutes, learning about each of their businesses.

When my energy levels started to drop and I felt the introverted excuses starting to kick in, telling me to go back to my room and be alone, I walked up to the bar and ordered a beer. I drank a third of the beer immediately and then slowly sipped the rest for the next hour and a half. There would be no excuses this conference — I wasn’t going to let my introverted habits win and force me back to my room when there were so many people around me that I could learn from.

I continued socializing, talking with others, and most importantly, listening to others for the entire rest of the reception. The value I got out of hearing other’s techniques, methods, and war stories was worth the price of the conference ticket alone.

The rest of the conference was a huge success.

Not only had I not let history repeat itself but I had broken out of my shell as well.

You get what you put into a conference

Since then I’ve attended too many conferences to count. Some of them great, some of them a borderline waste of my time.

But by in large I found there are two critical attributes in making a conference great:

  • You get what you put into it
  • Smaller, more intimate conferences are typically better

Just like college or practicing a sport/instrument, you get what you put into it.

If you skip class all the time and don’t make any friends, chances are you aren’t going to enjoy college.

Similarly, if you take up learning guitar and then only practice once a month for thirty minutes you can’t expect to get any better at it.

The same goes for a conference.

You need to walk into a conference with a list of what you want out of it. Go in there with goals, such as:

  1. I want to learn how to train my own custom object detectors from Davis King’s talk and workshop
  2. I want to ask Katherine Scott about satellite image analysis
  3. I want to ask Adrian for his advice on what it’s like starting a computer vision/deep learning company

Take the time to write out your goals of attending a conference before you even hop on the plane. This list will guide you and help ensure you make the most out of the experience.

Secondly, I’ve found that smaller, intimate conferences are typically better than larger ones.

There are too many people at large conferences (> 500-1,000 people). You can’t possibly meet them all — and the ones you do meet you can only chat with for 30 seconds before moving on to the next. This isn’t enough time to get to know the person, their projects, and how you might be able to help each other. These are missed connections. Connections that can have a big impact on your career and projects.

Furthermore, smaller conferences help you better plan (and achieve) your goals for attending the conference:

  • If there are a smaller number of overall attendees, the more time you’ll be able to fit in with the speakers and workshop hosts.
  • And with less overall attendees, the more intimate conversations you can have with the ones who attend.

Now, don’t get me wrong — you can’t run a poorly planned conference, including a less-than-adequate venue and subpar speakers and expect it to be better than NIPS.

But what you can do is craft a conference that includes the best of both worlds:

  • Highly renowned speakers
  • 1-on-1 time with experts in the field
  • Hands-on talks and workshops where you learn skills and techniques you can apply to your own datasets or projects that night
  • A kick-ass venue with lots of perks (beautiful hotel, views of the SF bay, catered lunch, evening events, open bars, walking distance countless, excellent restaurants in SF, etc.)

And that is exactly what PyImageConf is: small, intimate, and hands-on.

I created PyImageConf to be the conference I would want to attend (and you would too)

Figure 3: PyImageConf, the practical, hands-on computer vision and deep learning conference.

After attending many conferences, both in the computer vision/deep learning and entrepreneurial spaces, there are two must have characteristics that I look for:

  1. Small and intimate (< 250 attendees)
  2. Hands-on and practical — the ideas and algorithms discussed can be applied to your CV/ML (or business, in the case of entrepreneur conferences) that night in your hotel room

One of the worst feelings as a conference attendee is to feel like you’re just another statistic, lost in the crowd.

If you’ve ever attended a large conference before, you know it’s hard to make connections at these bigger venues. You either:

  1. Attend the conference with a group/make a group of friends quickly and hang out together the entire conference in order navigate the nearly overwhelming sea of attendees.
  2. Play a never-ending game of “speed dating” where you try to spend 30 seconds chatting with everyone. Not only will you be unable to chat with every person but (most importantly) 30 seconds is far too little for you to develop a lasting relationship.

Another aspect I love regarding small conferences is experts are more willing to share techniques they cannot share publicly, whether this is a new algorithm they are using to obtain higher accuracy than a competitor or a new method they cannot publish due to employer regulations or fear of ridicule. Experts, including myself, are more willing to be open and share in tiny groups.

And the techniques shared can have big impacts on your own projects.

I can’t tell you the number of times I’ve learned a new method at a conference and then applied to my technical work or business on the flight home from a conference (and was able to see the results quickly).

In general, the entrepreneur conferences I’ve attended excel at these practical tips and techniques. Larger computer vision and machine learning conferences have a lot to learn from this area.

Note: I’m making a generalization here to make a point, so please forgive me. Not all entrepreneur conferences are hands-on and practical (there are quite a few that are frankly awful, honestly). Similarly, not all large CV and ML conferences focus just on theory. But when you stick to the two characteristics I’ve suggested above, you’ll find that small entrepreneur conferences do a fantastic job with the tactical content.

The goal of PyImageConf is the blend the positive experiences I’ve had at both types of conferences and apply them to the computer vision and deep learning space.

You’ll learn real-world computer vision and deep learning techniques that you can then apply to your own datasets/projects that very evening (although I would suggest waiting until the flight home so you can make the most of the conference by creating lasting connections at the evening meet ups and open bar).

Why bother starting a conference?

Trust me, it’s not about the money.

Profit margins on conferences are ridiculously low, especially for ones that have capped attendee lists such as PyImageConf.

Unless your full-time job is to make a conference profitable by bringing in as many attendees as you possibly can, running a conference is not a way to make a long-term living. Any profit you do make (which is typically in the 10-20% range for small conferences) is completely and totally lost when you factor the amount of your own time spent planning and executing the conference.

At this point you’re probably wondering:

Adrian, if it’s not even economical for you to run a conference, why would you even bother? Why not invest your time into writing another book or putting together a new course?

The answer is simple:

It’s not about me, it’s about you, the PyImageSearch reader (and hopefully PyImageConf attendee).

PyImageSearch is more than just me.

Yes, I write the blog posts, books, and courses. But that doesn’t matter in the long-term and grand scheme of things.

What really matters is the community here at PyImageSearch.

We learn from each other.

We interact in the comments section.

We chat over email.

And PyImageSearch Gurus members share projects, algorithms, and techniques daily inside the forums.

Each and every day I learn something new from a reader’s question which required me to read up on a new technique, explore a dataset, or look at computer vision/deep learning problem from a different angle.

This is an incredible community and I believe it’s my responsibility to continue growing and nurturing it. While I may be the face of PyImageSearch I also humbly accept my role as the steward of the community — you’re the real PyImageSearch.

Simply put:

PyImageConf is about me facilitating the growth of the computer vision and deep learning community by creating a conference that hasn’t existed at this practical, hands-on level before.

Conferences, and the connections I’ve made at them, changed my life and this conference will do the same for you.

At PyImageConf I’ll make sure you meet at least one remarkable attendee before the conference starts so you’ll go in feeling like you’re part of the family rather than the shy outsider having to work up the courage to start a conversation.

I’d love to see you there and I hope you can make it.

Why should I attend PyImageConf?

You should attend PyImageConf if you:

  • Are an entrepreneur who is ready to to build the next computer vision or deep learning app
  • Are a student unsure of your career path, but ready to explore computer vision, deep learning, and AI
  • Are a computer vision hobbyist who loves building new projects and tools
  • Are eager to learn from the top computer vision and deep learning educators
  • Enjoy the teaching style of PyImageSearch and want personalized, live, in-person training

If this sounds like you, rest assured, this conference will be well worth of your investment of time, finances, and travel.

How do I know this?

Because I know there are two components to a conference such as this one:

  • The education, including talks, workshops, etc.
  • The connections made during the reception, between talks, and evening events.

For example:

At some point in the conference, you’ll need to get up and head outside to grab a cup of coffee, a snack, or get some fresh air. Perhaps this is during a talk you’re less interested in, the Q&A session, or right after a talk wraps up.

Just outside the doors to the ballroom you’ll also find a small number of other attendees doing the same. Some of these attendees will be engaged in a deep conversation regarding algorithms, techniques to apply to a dataset, or even their businesses/consulting work. Colloquially, this is known as the “hallway track” in conference lingo.

Some of the most valuable, lasting connections can be made during the “hallway track” as there are no distractions. It’s quiet. It’s intimate. It’s easy to strike up a conversation and learn what a small group of people around you are working on — and others will be happy to talk to you. Do not undervalue these conversations.

These connections made through these conversations enable:

  • You to meet someone who can help with your current project
  • Help you find your next job in computer vision or deep learning
  • Find a client for your current image processing business
  • Demo your latest project on your laptop, get advice, and continue to build your demo into an actual application
  • Find an attendee who is also struggling with grad school, but confident they want to pursue AI or ML

Here’s how I’ll make sure you get the most out of PyImageConf…

Figure 4: Before PyImageConf starts I’ll be creating a Slack group to help you network, create your list of goals, and find attendees you need to meet up with.

To help facilitate these connections, a couple months before PyImageConf 2018 I’ll be creating a Slack group for PyImageConf attendees.

This group will enable you to chat with other attendees before the conference is even underway. Use this time to learn what the other attendees are working on — and then use this information to help you craft your goal list I suggested above:

  • “Which speakers do I want 1-on-1 time with?”
  • “Which attendees can help with my current project?”
  • “Which attendees do I need to meetup with so I can demo my new project on my laptop?”
  • “Which group of attendees will I be grabbing dinner with so I can learn from their expertise?”
  • “Which attendees are hiring so I can land a new job in computer vision/deep learning?”
  • “Which attendees can give me advice on grad school?”

I’ll also be personally reaching out to each attendee and asking about your goals. If you don’t have a goals list I’ll help you put one together. And if you don’t know which attendees to chat with to achieve your goals, I’ll make sure to help you form a connection.

PyImageSearch is more than just a educational blog, it’s a community as well.

If you want to get to know readers in the community on a more personal level and meet with people that you can work with, consult with, and help you achieve your computer vision goals, then PyImageConf is the place to be.

Don’t miss out on tickets (they’ll likely sell out quickly). Make sure you join the early bird list now.

PyImageConf sounds great, what now?

In order to create a small, intimate conference, I’m capping the total number of attendees at 200.

Early bird tickets to PyImageConf 2018 go sale this Friday, January 19th. The general public sale will start Friday, January 26th.

At this point I’m not sure if there will be any tickets left by the time the general sale starts.

If you’re interested in attending PyImageConf and want a ticket, make sure you click the following link and join the early bird list:

Don’t miss a chance on your chance at a PyImageConf ticket, click here to join the early bird list.

For more information, please refer to:

If you have any other questions regarding the conference please use the comments section below or reach out via my contact form.

The conference is going to be a blast, I hope you can make it!

The post Why I started a computer vision and deep learning conference appeared first on PyImageSearch.



from PyImageSearch http://ift.tt/2qPZSKe
via IFTTT

No comments: