Closed pattacini closed 5 years ago
It came out that with the following parameters, the filtered depth gives us skeletons with the correct metrics:
[depth]
enable true
kernel-size 10
iterations 5
min-distance 0.5
max-distance 10.0
Skeletons with unfiltered depth: |
Skeletons with filtered depth: |
Thus, @vvasco could you please try out the options above within TUG experiments and report here any findings?
With the following parameters:
[depth]
enable true
kernel-size 4
iterations 3
min-distance 0.5
max-distance 10.0
I obtained the following result:
The ankles look reasonably close to the lines when I stand on them. I checked the metric I use for line crossing and it works out with these parameters.
Awesome! Let's go this way then ✨
Could you please put those parameters in master
for the TUG?
Sure!
Soon afterward, please close this issue with a pointer to the commit hash.
Done here c86e3b69e054939a6fa9b00498ded711acf04952
Once we have the skeletons retrieved from the filtered depth, we need to compensate for the incorrect depth by applying a kind of unwarping in order to adjust the skeleton pose.