Dual-Level Decoupled Transformer for Video Captioning

Video captioning aims to understand the spatio-temporal semantic concept of the video and generate descriptive sentences. The de-facto approach to this task dictates a text generator to learn from textitoffline-extracted motion or appearance features from textitpre-trained vision models. However, these methods may suffer from the so-called textbftextit"couple" drawbacks on both textitvideo spatio-temporal representation and textitsentence generation. For the former, textbftextit"couple" means learning spatio-temporal representation in a single model(3DCNN), resulting the problems named disconnection in task/pre-train domain and hard for end-to-end training. As for the latter, textbftextit"couple" means treating the generation of visual semantic and syntax-related words equally. To this end, we present D2 - a dual-level decoupled transformer pipeline to solve the above drawbacks: (i) for video spatio-temporal representation, we decouple the process of it into "first-spatial-then-temporal" paradigm, releasing the potential of using dedicated model(textite.g. image-text pre-training) to connect the pre-training and downstream tasks, and makes the entire model end-to-end trainable. (ii) for sentence generation, we propose Syntax-Aware Decoder to dynamically measure the contribution of visual semantic and syntax-related words. Extensive experiments on three widely-used benchmarks (MSVD, MSR-VTT and VATEX) have shown great potential of the proposed D2 and surpassed the previous methods by a large margin in the task of video captioning.