Low-resolution human pose estimation

Human pose estimation has achieved significant progress on images with high imaging resolution. However, low-resolution imagery data bring nontrivial challenges which are still under-studied. To fill this gap, we start investigating existing methods and reveal that the most dominant heatmap-based would suffer more severe model performance degradation from low-resolution, offset learning is an effective strategy. Established observation, in work propose a novel Confidence-Aware Learning (CAL) method further addresses two fundamental limitations of methods: inconsistent training testing, decoupled heatmap learning. Specifically, CAL selectively weighs respect to ground-truth confident prediction, whilst capturing statistical importance output mini-batch manner. Extensive experiments conducted COCO benchmark show our outperforms significantly state-of-the-art for human estimation.