Robust Pose Estimation in Crowded Scenes with Direct Pose-Level Inference Supplementary Materials

In Sec. 3 we present the overall objective of PINet, here we provide more details about the groundtruth Here we also follow [2] to take the maximum of confidence maps. Here we give the details of how to sample pose for PR refinement. We visualize some crowded scenes pose estimation results on both OCHuman and CrowdPose, as shown in Figure 1 and Figure 1.