ECCV 2018 Daily - Monday

Daily MONDAY 10 SiVL (Shortcomings in Vision and Language) workshop was held on Saturday. We weren't there, but the organizers were kind enough to tell us in detail what happened there. The workshop brought together experts at the intersection of vision and language to discuss modern approaches, tasks, datasets, and evaluation metrics for significant problems on the integration of these two modalities. The aim of the workshop was to facilitate discussion of novel research directions and to steer the community towards high- level challenges affecting the vision and language community broadly. Inspiring talks were given by the invited speakers. Not surprisingly, Neural Networks dominated the scene, but interestingly the popular end-to-end one-big-black-box architecture has, in many case, been replaced by a more modular one. On the one hand, traditional Computational Linguistics components -- such as Part-of-Speech tags, question type detection etc. -- have found their role in the NN architecture, and the importance to carry out qualitative analysis instead of only quantitative comparison of models’ performance has been highlighted (Aishwarya Agrawal). On the other hand, traditional Computer Vision components -- such as object localization -- have been put back at work into end-to-end structures, showing the need for vision and language systems to focus on entities (Lucia Specia). Other interesting issues that emerged are the need to develop models that are not task specific and furthermore are able to deal with dataset bias (Vicente Ordóñez Román.) An example of the social impact of this research line has been shown by Danna Gurari, who presented her project to develop models to assist blind people. The workshop hosted the organizers of the 1st Visual Dialog Challenge , who introduced a new robust evaluation metric for Visual Dialog and gave a deep quantitative and qualitative overview of the systems that participated in the competition. Complete slides with challenge overview, standings, analysis available here . All the slides presented during the day are available from the workshop web- site Shortcomings in Vision and Language 12 Workshop: SiVL