Thank you for the great paper and open-sourcing your work. I am trying to train the network on a costume dataset that contains multiple people performing different tasks in each frame. I am only interested in a few activities but there are people doing other stuff in the background as well. I was wondering if I need to annotate all actors or if I can just annotate the ones that I'm interested in? I felt like not annotating other actors has a negative effect on actor detection part. I thought about annotating them all in one class as "backgroud" but that seems to be confusing for the network as well since they are doing different activities. I noticed one of the GIFs with people fencing contains a guy siting on a chair in the background and he is not detected, is this the same situation as mine? I'm not sure if UCF dataset contains "siting" class.
sorry for the long question.
Thank you for the great paper and open-sourcing your work. I am trying to train the network on a costume dataset that contains multiple people performing different tasks in each frame. I am only interested in a few activities but there are people doing other stuff in the background as well. I was wondering if I need to annotate all actors or if I can just annotate the ones that I'm interested in? I felt like not annotating other actors has a negative effect on actor detection part. I thought about annotating them all in one class as "backgroud" but that seems to be confusing for the network as well since they are doing different activities. I noticed one of the GIFs with people fencing contains a guy siting on a chair in the background and he is not detected, is this the same situation as mine? I'm not sure if UCF dataset contains "siting" class. sorry for the long question.