Abstract: Multilabel image recognition (MLR) aims to annotate an image with comprehensive labels and suffers from object occlusion or small object sizes within images. Although the existing works ...