The invention concerns a method and a device for accessing a spatio-temporal part of a video sequence of images compressed using a coding format implementing a prediction.The method comprises the steps of: defining a target spatial region corresponding to said spatio-temporal part to access; obtaining compressed data corresponding to a selection spatial region in at least one image of the video sequence, said selection spatial region including said target spatial region; obtaining at least one item of updating information for updating the selection spatial region that is provided for selecting compressed data in a following image of the video sequence, said updating information depending on the decoding of the compressed data corresponding to said target spatial region.