HandyPose: Multi-level framework for hand pose estimation

Hand pose estimation is a challenging task due to the large number of degrees freedom and frequent occlusions joints. To address these challenges, we propose HandyPose, single-pass, end-to-end trainable architecture for 2D hand using single RGB image as input. Adopting an encoder-decoder framework with multi-level features, along novel waterfall atrous spatial pooling module multi-scale representations, our method achieves high accuracy in while maintaining manageable size complexity modularity network. HandyPose takes approach representing context by incorporating information at various levels network mitigate loss resolution pooling. Our advanced leverages efficiency progressive cascade filtering larger fields-of-view through concatenation features from different module. The decoder incorporates both generation accurate joint heatmaps stage. results demonstrate state-of-the-art performance on popular datasets show that robust efficient estimation.