DeepLabCut feature extractor

[julianzille]

Hi everyone.

I was hoping someone might be able to help me with this: does the DeepLabCut feature extractor used (with ResNet152 backbone) estimate the cheetah's pose from an entire frame, or is the body first localised by calculating its bounding box, before performing feature extraction?

Thanks in advance :)

[MMathisLab]

hi @julianzille from the entire frame!

[julianzille]

Thank you @MMathisLab. I was also wondering if the code that was used to calculate the RMSE and s.e.m of the DLC feature extractor is available? Otherwise, could you perhaps share how it was calculated? Ie. is it taken from the average euclidean error across all keypoint predictions, or across the average rmse across all frames (for example)?