Deep Attention Based Semi-supervised 2D-Pose Estimation for Surgical Instruments

For many practical problems and applications, it is not feasible to create a vast accurately labeled dataset, which restricts the application of deep learning in areas. Semi-supervised algorithms intend improve performance by also leveraging unlabeled data. This very valuable for 2D-pose estimation task where data labeling requires substantial time subject noise. work aims investigate if semi-supervised techniques can achieve acceptable level that makes using these during training justifiable. To this end, lightweight network architecture introduced mean teacher, virtual adversarial pseudo-labeling are evaluated on surgical instruments. applicability pseudo-labelling algorithm, we propose novel confidence measure, total variation. Experimental results show utilization improves unseen geometries drastically while maintaining high accuracy seen geometries. RMIT benchmark, our outperforms state-of-the-art with supervised learning. Endovis algorithm baseline achieving new performance.