A Neural Multi-sequence Alignment TeCHnique (NeuMATCH)

2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition(2018)

引用 8|浏览34
暂无评分
摘要
The alignment of heterogeneous sequential data (video to text) is an important and challenging problem. Standard techniques for this task, including Dynamic Time Warping (DTW) and Conditional Random Fields (CRFs), suffer from inherent drawbacks. Mainly, the Markov assumption implies that, given the immediate past, future alignment decisions are independent of further history. The separation between similarity computation and alignment decision also prevents end-to-end training. In this paper, we propose an end-to-end neural architecture where alignment actions are implemented as moving data between stacks of Long Short-term Memory (LSTM) blocks. This flexible architecture supports a large variety of alignment tasks, including one-to-one, one-to-many, skipping unmatched elements, and (with extensions) non-monotonic alignment. Extensive experiments on semi-synthetic and real datasets show that our algorithm outperforms state-of-the-art baselines.
更多
查看译文
关键词
NeuMATCH,heterogeneous sequential data,video to text,inherent drawbacks,Markov assumption,future alignment decisions,similarity computation,end-to-end training,end-to-end neural architecture,alignment actions,flexible architecture,alignment tasks,nonmonotonic alignment,dynamic time warping,conditional random fields,long short-term memory blocks,neural multisequence alignment technique,LSTM,CRF,DTW
AI 理解论文
溯源树
样例
生成溯源树,研究论文发展脉络
Chat Paper
正在生成论文摘要