Parallel-fusion LSTM with synchronous semantic and visual information for image captioning
Abstract: Highlights•A novel parallel-fusion LSTM is proposed for image captioning.•We propose two structures pLSTM-A and pLSTM-G according to fusion strategy.•pLSTM-A pays attention to the crucial information at different time step.•Each gate of visual LSTM is guided by synchronous semantic attribute in pLSTM-G.•Experimental results prove that our model performs state of the art methods.
Loading