search

Embracing / Active3DPose

Official implementation of the ICLR2023 paper "Proactive Multi-Camera Collaboration for 3D Human Pose Estimation"
Apache License 2.0
46 stars 4 forks source link

Custom camera input #4

Open turtlebot1 opened 6 months ago

turtlebot1 commented 6 months ago

How to use custom camera input and process them instead of using the simulation output of unreal engine? Please let me know about it. Thank you

Embracing commented 6 months ago

We define the action space of cameras here. It is actually a matrix of [C, 5] with values 0-2. C is the number of cameras. 5 is the dimensions to take action. Please see this.

Take a single camera as example. It can move along xyz and rotate with pitch and yaw angle. If you set an action as [0, 1, 2, 0, 1], it will move along x and z-axis with a step of -30 and 30, and rotate with pitch angle of -3 degree. Stride is defined here.

If you would like to use custom camera input, you may need to modify the stride to fit your custom camera's config and only use the action dimensions where your camera is able to take an action.

turtlebot1 commented 6 months ago

Thank you very much for your valuable assistance. For my specific use-case, I am particularly interested in the 'Perception Module' as outlined in your manuscript. Could you please guide me to where I can find the relevant code for this module? Any assistance in this regard would be greatly appreciated.

Thank you once again!