una-dinosauria/3d-pose-baseline

3d_data.json

Closed this issue · 9 comments

AndroidUnityGit commented

@una-dinosauria

  1. operating system: win10
  2. tensorflow version :1.4
  3. python version :3.7
    run python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory --write_gif --gif_fps 24
    Used Pre-trained model .

Generated 3d_data.json and animation.gif
3d_data.zip
animation.zip

but Why are there 32 points in 3d_data.json? How do these points correspond to joints?

liuzitongtt commented

I have the same question ？ Which of these 32 coordinate positions are valid?

liuzitongtt commented

I have found this correspondence in the code

ArashHosseini commented

Hi @AndroidUnityGit ,
can you please provide me with the following information and i will try address this issue this weekend:

  1. what are you using to creating the 2d screen space skeleton prediction? (openpose etc. ...)
  2. tell me the version of "it" (point 1)
  3. upload the output json file of "it" (point 1)

Thx

AndroidUnityGit commented

@ArashHosseini Thank you for your reply!

  1. I used openpose to create the 2d screen space skeleton prediction

2.I used OpenposeDemo openpose-1.5.1-binaries-win64-only_cpu-python-flir-3d.zip

https://github.com/CMU-Perceptual-Computing-Lab/openpose/releases/download/v1.5.1/openpose-1.5.1-binaries-win64-only_cpu-python-flir-3d.zip

2d_Screen_prediction.zip

3d-pose-baseline Used yours fork
Generate 3d_data.json
3d_data.zip

ArashHosseini commented

@AndroidUnityGit thanks for your answer, now i have to ask you for some patience, i will investigate this behavior and update you until monday. thx

timtensor commented

@AndroidUnityGit , i could not find it sorry. could you point the joint locators as well ?

ArashHosseini commented

Hi @AndroidUnityGit,
thank you for your patience. Your code works as it should, this behavior is expected. Now to your question, the maya_skeleton.py is handling the mapping of the 32 joints, see. how ever i visualized the hierarchy for better understanding, see. The number in "jnt_n_threed" corresponds to the output frame array in "3d_data.json".