Build what you need and use what you build

Build what you need and use what you build. This is a core philosophy of my research. It shifts the focus away from writing and publishing “papers” to what really matters — impact.

At the start of any research project, my students know that I will ask them “Who’s your customer?” By customer, I don’t mean “paying customer”. I mean “who needs what you’re proposing?” Who will use it? Why do they need it? If you can’t answer this, then your work is likely to be irrelevant.

A good answer to “who’s your customer?” can be “me”. If you need it, then you need it. And if you need it, there are probably other people out there in the world like you who will need it too. Corollary: if you aren’t going to use it, why do you think others will?

Importantly, if you're your own customer, then it doesn’t matter what Reviewer 2 thinks. It doesn’t matter if it gets published. You need it and will build it regardless of what the world thinks. It takes you out of the publishing rat race and focuses you on what you think is important.

An example is BEDLAM - our synthetic dataset for training 3D human pose and shape regressors. When we were working on BEDLAM, I said to the team “it doesn’t matter if this gets published because we need it.” We were going to build it no matter what. It didn’t exist and we needed it. We were our own customers. (It got published at CVPR 2023, precisely because it is useful and we demonstrated this.)

The second part of my maxim is to “use what you build.” This is critical. If you write a paper and never build on it, then why do you think others will? Papers that you don’t build on are dead ends. Cute ideas with no impact. These papers don’t change the field.

If the idea is good, one paper isn’t enough. People don’t pick up on new ideas right away. They don’t see the future you see… yet. You need to teach the future. If you have a new idea, you will have to show people why it’s also a good idea. You teach the future by demonstration.

An example: we introduced the idea of using parametric 3D human shape models for human pose estimation and tracking at CVPR 2007. At the time everyone was estimating 3D stick figures and they kept on doing so for many years. The impact was modest initially. I knew it was a good idea for many reasons but the world was busy with its own ideas.

So we built and built and built on top of this idea. We showed how you could leverage 3D humans to do more, do better, and solve new problems. It took many papers for us to teach what we already saw. It took making software and data available to enable others to follow. You teach by example.

In my department, we build what we need and we use what we build. This focuses us on impact not papers. It helps us build a community of “customers”. Impact comes from a community that expands on your ideas. Teach your ideas. Provide the tools for others to follow. And then be happy when others surpass you. That's impact.