search

African-Robotics-Unit / AcinoSet

Camera calibration & 3D pose estimation tools for AcinoSet
https://www.dropbox.com/sh/mto9ilgnopeixyp/AABSelc43yid-yenPVbeU9Ioa?dl=0
58 stars 14 forks source link

DeepLabCut feature extractor #19

Open julianzille opened 1 year ago

julianzille commented 1 year ago

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 commented 1 year ago

hi @julianzille from the entire frame!

julianzille commented 1 year ago

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)?