Including skeleton in training with single-animal DLC (2.1.9)

Usage & Issues
#1

Hi all,

I was digging through pose_cfg.yaml files in train subfolders of dlc-models and noticed that maDLC uses the following two parameters:

partaffinityfield_graph:
- - 0
  - 1
- - 0
  - 2
- - 0
  - 3
...
partaffinityfield_predict: true

Am I correct in assuming this is how the overconnected skeleton is included into training? If so, could the skeleton also be used in training of single animal networks (DLC 2.1.9) by just including these two parameters to appropriate pose_cfg.yaml files, or is it more involved than that?

Many thanks for your insights!

1 Like
#2

It would be a bit more involved than that. In particular, the structure of the training data is slightly different and a few more parameters are set. So the easiest way is to select the multi animal version during training set creation.

Also maDLC is only available with v2.2.