What's a deep learning project that you'd like to do if only you had time?

There’s so much to learn out there and only so few hours in the day…

Not only that I feel like my own space of ideas for projects is limited by my own interests, what I know, and what I come across.

I think a good way to expand my world would be to hear what projects people would love to work on…if only they had the time. Would love to hear from @trust_level_1 on this

Oh man, where to start :sweat_smile:

If I had the time, I would definitely love to work on developing a 3D-Pose estimation model. This could be used as a relatively low-cost motion capture system - so you could animate video game characters without having to invest in expensive motion capture hardware/software. Not to mention the applications in rehabilitation monitoring for athletes/patients.

Current 2D pose estimation models are good, but there are still issues with occluded joints, which leads to inaccurate/noisy results when the body is in certain positions.

As a workaround, I’ve had a project in the back burner where I was thinking of using 2 camera inputs - each being processed by a 2D pose estimation algorithm - and then trying to triangulate the 3D pose using trigonometry. Maybe one day :smile:

1 Like

That’s one of the biggest trends I picked up on at the ECCV conference: pose estimation.

There was a lot of posters presented on that topic.

Here are a couple from there: