Semantic Kitti [for normal people]
The purpose of this post is to see if there could be some interest or if what I’m thinking resonates in my circle of influence.
Warning! The content might be too lengthy or technical, but I’ll try to make it interesting.
A new set of data called “Semantic Kitti” (http://semantic-kitti.org/) has been released. It is an important dataset useful for developing Self-driving car perception technology. If you don’t know what that means, “Wollup” for a second, I’ll explain.
So, a self-driving car is simply a car that can drive by itself. Simple! But for the car to drive by itself, it needs to be able to “see” the environment. There are three ways that are (presently) possible for the car to “see” (Notice I used “see” because it’s not exactly the way we see. But for better understanding let’s agree they mean the same thing). The three ways are; through normal images (like the one you take on your smartphone), by radio waves called Radar and by laser light (hmmm!) also called Lidar. (Please accept that there is something called Radar and Lidar for now. :) ). All these three methods are useful for different purposes. Images are best for “seeing” colours, textures, objects, generally things humans can see through the eyes. Radar is good for seeing how fast things are moving. Speed. And Lidar is good for seeing where things are. Location or distance. It turns out that knowing where things are is very important for self-driving cars because you don’t want your car to run into other things (believe me, it is usually bad when this happens). Images can be used to “see” distance, but we need to do more mathematics to achieve the same results (many people don’t like mathematics, I know. So we stick with Lidars even though they are more costly. Elon!). Remember the dataset I said was released? Yeah, that is Lidar. Why was it released? In short, it is good for research (and humankind!?). Particularly, when “Kitti”, the group that first collected the data released their version, it helped many smart people (aka, researchers) understand and develop algorithms (or solutions) that enable the car to drive more safely. This problem, however, is very challenging and we need to modify the data in many ways in other to make it more useful (Yes, I’m part of the smart people. Haha!). Even though the original dataset has been released since 2012, we are still using it to discover new things in 2019.
Now to “Semantic Kitti”, which is another modification of the original dataset. This version has been carefully and manually edited and crafted to not only allow us to know the location of objects in the scene but also the kind of objects they are. Hence the word semantic. So you see how important this is. Finally, to what I have been driving at. I’d like to play with this dataset, for fun, and for research. And I just wonder if someone else would be interested.
More interestingly, there is a competition to use this dataset to accomplish some cutting edge tasks. I think there are real opportunities here. Most especially for AI enthusiasts. It does not matter if you do not understand everything at this point. What you need is the willingness to do some real work. Be able to learn fast. Know some mathematics and python programming and git. I think I’m saying too much already but these are important too.
Summarily, I am just putting this out there to see reactions.
This actually turns out to be a good research area (or project topic) for computer science, computer, mechanical and electrical engineering. So share with anyone you know in these areas. I would gladly mentor if need be.
Please note that all these might not happen. The reason for this post is to see if it can.
I need to stop now.. :)
Hello! Before you go. Ask that question in the comment. I’ll answer.
;)
Note: This post was first created for Facebook, hence the way it was written. I hope you find it fun reading it here too.
