This feature extraction is in line with the recursive feature elimination script used to optimize DLStream + SiMBA classifiers.
It is adapted from the original simba/features_scripts/extract_features_14bp.py and should fit into the SiMBA workflow without issues. Naming is consistent with previous feature_scripts.
Classifiers trained with this feature selection can be used with the optimized feature extraction script (by Simon) in DLStream that pushes classfication time down to sub pose estimation inference speed. A huge benefit. It was already used for anogenital approach classification. See DLStream Code and DLStream Wiki.
Hopefully this will enable the community to directly train "real-time" optimized classification in SiMBA and implement them into DLStream. Let me know what you think.
This feature extraction is in line with the recursive feature elimination script used to optimize DLStream + SiMBA classifiers. It is adapted from the original simba/features_scripts/extract_features_14bp.py and should fit into the SiMBA workflow without issues. Naming is consistent with previous feature_scripts.
Classifiers trained with this feature selection can be used with the optimized feature extraction script (by Simon) in DLStream that pushes classfication time down to sub pose estimation inference speed. A huge benefit. It was already used for anogenital approach classification. See DLStream Code and DLStream Wiki.
Hopefully this will enable the community to directly train "real-time" optimized classification in SiMBA and implement them into DLStream. Let me know what you think.