Document worth reading: “Video Description: A Survey of Methods, Datasets and Evaluation Metrics”

Automatic video description is useful for serving to the visually impaired, human computer interaction, robotics and video indexing. The last few years have seen a surge of evaluation curiosity on this area as a result of of the unprecedented success of deep finding out in computer imaginative and prescient and pure language processing. Numerous methods, datasets and evaluation measures have been proposed inside the literature calling the need for an entire survey to greater focus evaluation efforts on this flourishing course. This paper options exactly to this need by surveying state of the paintings approaches along with deep finding out fashions; evaluating benchmark datasets in phrases of their space, amount of programs, and repository dimension; and determining the professionals and cons of various evaluation metrics equal to BLEU, ROUGE, METEOR, CIDEr, SPICE and WMD. Our survey reveals that video description evaluation has a protracted strategy to go sooner than it could probably match human effectivity and that the precept causes for this shortfall are twofold. Firstly, current datasets do not adequately signify the range in open space films and difficult linguistic constructions. Secondly, current measures of evaluation shouldn’t aligned with human judgement. For occasion, the an identical video can have very utterly totally different, however proper descriptions. We conclude that there is a need for enchancment in evaluation measures along with datasets in phrases of dimension, vary and annotation accuracy consequently of they instantly have an effect on the occasion of greater video description fashions. From an algorithmic stage of view, evaluation of the define top quality is troublesome consequently of of the difficultly to judge the extent of contribution from seen choices as compared with the bias that comes naturally from the language model adopted. Video Description: A Survey of Methods, Datasets and Evaluation Metrics