Enable demo (inference on multi-view images)

[kristijanbartol]

Motivation

The most natural expectation is to have a demo script to run results on your own data, which is the main "selling point" of the paper. To make it more convenient for the users, it would be nice to have a script that directly takes images and produces 3D poses, without having to go to other repository first to produce 2D keypoint estimations.

Description

In order to support demo, several steps need to be prepared:

• 2D keypoint detections (using a "custom" detector, such as OpenPose)
• camera parameters estimation (in case the parameters are not already obtained)
• create a custom dataset on-the-fly
• 3D keypoint estimations similar to the current test() function

[anas-zafar]

Hi @kristijanbartol! Thanks for the amazing work. Any updates on v0.2?

[kristijanbartol]

Hi @anas-zafar , thank you for the reminder and I'm sorry to disappoint! I am currently working hard to finish my new work so I can't finish v0.2. As a minor excuse, most of the v0.2 is out of the scope of the paper (it is moving towards future work), so I would be glad to assist you with v0.2 for your use case or, ideally, assist you with the pull request to update this repository. :) I am planning to go back to this project by the end of May!