A method includes receiving, by a processing device of a content sharing platform, a video content, selecting at least one video frame from the video content, subsampling the at least one video frame to generate a first representation of the at least one video frame, selecting a sub-region of the at least one video frame to generate a second representation of the at least one video frame, and applying a convolutional neuron network to the first and second representations of the at least one video frame to generate an annotation for the video content.