Instrument detection and pose estimation with rigid part mixtures model in video-assisted surgeries.

•A method for 2D pose estimation of multiple non-rigid and robotic instruments is proposed.•Rigidly structured model expresses deformations of surgical instruments by encoding diverse, pose-specific appearance mixtures.•Extensive diagnostic experiments inform feature regularization significantly improves learning articulated tool models from videos.•The method successfully leverages the shaft part that improves estimation of endeffector articulation.•The algorithm obtains state-of-the art results on publically available datasets of nonrigid tools.