Self-Guiding Multimodal LSTM—When We Do Not Have a Perfect Training Dataset for Image Captioning | IEEE Journals & Magazine | IEEE Xplore