Show, Deconfound and Tell: Image Captioning with Causal Inference | IEEE Conference Publication | IEEE Xplore