Interact-Pose Datasets for 2D Human Pose Estimation in Multi-person Interaction Scene

In recent years, several excellent works dealing with human pose estimation in complex multi-person scenes have focus on the problems of complex backgrounds and the multiplayer scene. However, when facing the scene of multi-person interaction, the results of current mainstream algorithms are still unsatisfactory and some common datasets are not suitable for coping with interaction problems. Therefore, we propose a new dataset named Interact-Pose for solving multi-person interactions problems. Firstly, We use the MSCOCO format to annotate Interact-Pose. Except that, we adopt the corresponding data augmentation scheme to exchange the background of the Interact-Pose Dataset to make it more complex and have better generalization performance. Then it is trained after being fused with the MSCOCO dataset. After training on HigherHRNet, the average AP value of our test results is 67.3% on the Validation set of COCO2017, which is higher than that of the test only being trained by MSCOCO.