Appreciating Human Vision

Before this project, I was already aware that the human visual system is very powerful. At a quick glance, you can tell the gist of a scene - what the scene is about (Potter et al., 2014). You could even describe some of the scene's properties - how open it is, whether it is a natural or an urban setting, or whether it is an indoor or an outdoor scene (Boucart et al., 2013). Even when the scene is packed, you seem to be able to find what you're looking for (Wolfe et al., 2011), like a chocolate bar in a supermarket shelf, your phone on a cluttered desk, or Waldo.

A great example of the human visual system's power is in detecting cancers. Mammographers can detect the presence of cancer with above chance accuracy even though they were only shown the scan for 500ms. They can do this even if the mammogram is of a breast that will develop cancer three years later, or if the breast shown is opposite to the one with cancer (Brennan et al., 2018)

That's not to say that the human visual system is impenetrable though. Humans are often vulnerable to optical illusions, such as the Hermann Grid Illusion (Schiller & Carvey, 2005). Flip a face upside-down and it is more difficult to identify it (Yin, 1969). When given chest CT scans to detect lung nodules in, 20 out of 24 expert radiologists failed to spot a photoshopped gorilla in the last scan (Drew, Vo & Wolfe, 2013).

The human visual system may be capable of handling vast amounts of information, but it is still subject to constraints that seem like a hindrance on the surface but are actually used to guide behaviour accurately and to achieve ones goals successfully.

Still, the human visual system is truly remarkable and my curiosity on it has only grown throughout the time I have been studying it.

Starting the Project

To begin the project, I needed a set of images - the images had to be of naturalistic scenes with a certain object within it and this object had to have a characteristic sound. For example, a kettle has a characteristic ‘whistling’ sound and tends to be found on kitchen counters. Kettles aren’t usually found on top of a car parked on the side of the street.

Once I had these images, as well as recordings of their characteristic sounds, the next step was coding the experiment on MATLAB.

The experiment was a speeded object localisation task, where participants would be presented with one of these scenes and asked to indicate whether an object was left or right of the image. Sometimes characteristic sounds would accompany the image. Very simple.

Starting the project also involved piloting, which is good practice for any behavioural experiment. This means going through the entire experiment yourself (or asking other people to) to get an idea for the ‘feel’ of the experiment: not too hard, not to easy. You can make sure the data gets recorded properly, stimuli are presented as they should be and get a glimpse for what the data could look like.  'A practical guide for studying human behaviour in the lab' (Barbosa et al., 2022) is an excellent guide on good practices for behavioural studies which I highly recommend.

Not too hard, not too easy

It was just my luck (or perhaps my inexperience) that I encountered both of these problems during piloting.

At first, the task was too easy. I should have realised earlier that - of course - humans can find objects even from just a brief glimpse.

I made a few changes. Then, the task was too hard. Of course - if there are too many things in the picture, or if the object is too small, humans struggle with just a brief glimpse.

The problem is that very easy tasks lead to 'floor performance' and very hard tasks lead to 'ceiling performance'. Both of these could hide the effect I'm trying to find so my efforts further down the line might be wasted.

It was a nightmare trying to find the right balance, and I didn’t expect that coding an experiment would be the easier than finding images. Coding a working version of the experiment took around about an hour, the same amount of time to find sixteen good images or photoshop six. Just doing the experiment only takes 25 minutes to complete.

I had to learn how to photoshop in an hour (I'm still bad at it). I also easily lost track of which images I have used. But - worst of all - I'm now delayed in the project. I was meant to begin data collection two weeks ago, but I have only recently begun. And the people I wanted to ask to take part have now also gone home for the summer, so it has been more difficult finding participants.

As frustrating as this was, it has also been thrilling to see a project on the verge of collapse. Exaggeration now, but back then a reality.

A learning experience

As I frantically try to put a dog on a couch, I had a thought. If I had a sense for how these images would be like in the first place, this wouldn’t have been a problem. My supervisor, for example, has a great sense of working out how hard a task is. I can't say I have any regrets, however, because - of course - I will learn from and carry this experience with me from now on.

In terms of my leadership development, this was a real eye-opener in how acknowledgment of people’s experiences is crucial. Really, it's a shortcut for problem solving, as well as a method for improving this skill overall. Though I wasn't necessarily leading a team, this was a further demonstration of why teamwork and learning about oneself is valuable.

From the perspective of research, I also got a sense for how limited available datasets are for purposes like this. Good quality stimuli are hard to come by. There is a lot effort that tends to be unnoticeable once the experiment is ready to go that I have really, really come to appreciate.

Last but not least, this situation was a vivid reminder of how I enjoy studying neuroscience in all its wonders - from aspects of design, research and application. In such a short amount of time, I had never experienced feelings of excitement, frustration and motivation. In short, I learnt that it was not mistake that I underestimated the human visual system.