More infomation and nice pictures can be found at the horse-10 site!
- download the data and unzip:
THe data is available from: https://huggingface.co/datasets/mwmathis/Horse-30
tar -xvf horse10.tar.xz
- Train your model for the three splits!
The downloaded data is in DeepLabCut format with information regarding the splits stored in "/training-datasets/iteration-0/UnaugmentedDataSet_HorsesMay8".
- Evaluate each split. To normalize the different horse sizes, we normalize to the nose-eye distance for each horse. Here is some example code as well as the relevant distances are stored in this file.
For more details check out the paper: Pretraining boosts out-of-domain robustness for pose estimation published in the IEEE Winter Conference on Applications of Computer Vision 2021.