Related to #64 but for joint IDs instead of camera IDs
I'm interested in this specifically for when we do amputation experiments: removing some part of one or more of a fly's legs, and then wanting to do 3D pose estimation on the remaining parts of the fly. The 2D pose estimation gets confused by the missing bit of the leg, which isn't ideal but not worth fixing because it would require manual labeling and retraining DeepFly2D - and ultimately local errors are fine as long as we can have the 3D triangulation and video making steps ignore the terrible 2D estimations for the missing joints.
I specifically would like the following to work - this includes a path to a real dataset that I have, which I'm happy to provide for testing this feature:
cd /mnt/upramdya_data/JSP/SCAPE/250721_IN14B5x8m_invivo/Fly3/006_T1/
df3d-cli behData --ignore-joints 3 4 --video-2d --video-3d
where the joint ID values 3 and 4 come from the standard list of joint IDs:
Currently deepfly3d 2d output on the dataset I refer to above looks something like this:
Related to #64 but for joint IDs instead of camera IDs
I'm interested in this specifically for when we do amputation experiments: removing some part of one or more of a fly's legs, and then wanting to do 3D pose estimation on the remaining parts of the fly. The 2D pose estimation gets confused by the missing bit of the leg, which isn't ideal but not worth fixing because it would require manual labeling and retraining DeepFly2D - and ultimately local errors are fine as long as we can have the 3D triangulation and video making steps ignore the terrible 2D estimations for the missing joints.
I specifically would like the following to work - this includes a path to a real dataset that I have, which I'm happy to provide for testing this feature:
where the joint ID values 3 and 4 come from the standard list of joint IDs:
Currently deepfly3d 2d output on the dataset I refer to above looks something like this: