una-dinosauria/3d-pose-baseline

No 3dpose visualization in demo

tiffany0117 opened this issue · 7 comments

tiffany0117 commented

Hi,
Thanks for your wonderful works, but when I try the code in README for quick demo, there is no skeleton of 3dpose in visualization, and I found something strange at the end of training.

  1. Operating system : windows 10

  2. Tensorflow version : 2.3.0

  3. Python version : 3.8.8

  4. The stack trace of the error that you see

    =============================
    Global step: 24371
    Learning rate: 9.90e-04
    Train loss avg: 0.1195
    =============================
    ===Action=== ==mm==
    Directions nan
    Discussion nan
    Eating nan
    Greeting nan
    Phoning nan
    Photo nan
    Posing nan
    Purchases nan
    Sitting nan
    SittingDown nan
    Smoking nan
    Waiting nan
    WalkDog nan
    Walking nan
    WalkTogether nan
    Average nan
    ===================
    Saving the model... done in 1979.55 ms

image

fantasy870727 commented

I have same problem.Help me,please!

shin-ka commented

@tiffany0117 were you able to solve the problem?

tiffany0117 commented

@shin-ka I found that this problem is caused by the result of the predicted joints are not those we want like below :
#77 (comment)
the result are the other 16 static joints.
However, I have no idea how to solve this problem.

una-dinosauria commented

It is likely that a library we depend on changed and it's affecting the results. I'd recommend trying with older version of pytorch. I'll try to put together a docker image for this as well.

shin-ka commented

@una-dinosauria thanks for the suggestion! I downgraded to tensorflow 1.12 and it works.
@tiffany0117 fyi

una-dinosauria commented

Sorry, I meant tensorflow, not pytorch. Thanks for finding the fix @shin-ka!