Recognizing Human-Object Interactions in Still Images by Modeling the Mutual Context of Objects and Human Poses