weigq/3d_pose_baseline_pytorch

How to visualize outputs into images?

Opened this issue · 3 comments

ZHONGCHUYUN commented

Hi, I am totally new in this field, if this is a stupid question plz forgive me. I was wondering how to visualize the results. Do we need to add some codes on main.py like 'cv2.imshow'? And why the shape of inputs (64, 32) is not the same as the shapes of both targets' and outputs' (64, 48). Thank you in advance.

flyyyyer commented

Could you provide the code of visualize the outputs?

alvarogonjim commented

@ZHONGCHUYUN You need to define the skeleton, which joint correspond to which bone, you can see in the original work this information. The output is normalized so you need to unnormalized first in order to visualize the skeleton, you can either create your own code with matplotlib or use the original work for that. To answer your second question:

  • The shape of the input (64, 32) is the 2D ground truth. 64 is the batch size that you are using, then 32 corresponds to 16 joints * 2 = 32

  • The shape of the output (64, 48) is the 3D prediction, 64 again is the batch size. Then 16 joints * 3 = 48