Sequence-to-Sequence Video Captioning with Residual Connected Gated Recurrent Units

Recurrent neural networks have recently emerged as a useful tool in computer vision and language modeling tasks such image video captioning. The main limitation of these is preserving the gradient flow network gets deeper. We propose captioning approach that utilizes residual connections to overcome this maintain by carrying information through layers from bottom top with additive features. experimental evaluations on MSVD dataset indicate proposed achieves accurate caption generation compared state-of-the-art results. In addition, integrated our custom-designed Android application, WeCapV2, capable generating captions without an internet connection.