Can the primary regions in an image be automatically generated?

[gunners1886]

Hello, I found the primary regions you use in the paper Contextual Action Recognition with R*CNN is annotated data. Is there some method that can automatically generate the primary regions in an input image? if not, how can we recognize the actions in an input image without annotation information?

[gkioxari]

For the tasks tackled in our paper, the primary region always refers to the person that is being classified. Usually the boxes of the people are provided (in order to disentangle the task from the task of person detection). However, I do agree that in practice you don't have those boxes. If that is your case, you should probably run a person detector on your dataset and get the highest scoring activations to be the primary regions.