You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I managed to fork it and remove that dependency by using some meta information that was available in mmpose.
I did a few tests with in the wild videos. Despite PAFUSE using temporal information, the length of the bones had a lot of variations over time, sometimes up to 2x. For example, the upper arm to forearm ratio could vary over time up to 2x. These of course happen on poses where the arm comes at a high angle and most of the length is from the depth.
Did you observe this also for in the wild videos? This seems strange since not only these ratios should be fixed by tracking the same person, but also this ratio doesn't vary that much between different persons.
Is the H3WB dataset a hard requirement to just test your in-the-wild example?
From the code it seems that it is.
Ça aurait été gentil de le mentionner.
The text was updated successfully, but these errors were encountered: