A method for composing an intermediate representation of a video sequence from at least two different representations of the video sequence, the at least two representations being spatially partitioned into a plurality of tiles, each tile being associated with compressed data in the representation. The method is such that the tiles can be decoded independently from one another and at a given time the various representations are partitioned identically. The method includes selecting, according to at least one selection criterion, a set of tiles forming a partition of the sequence from the at least two representations, and composing an intermediate representation of the video sequence from compressed data associated with the selected set of tiles. The method can be implemented in a video decoding terminal or broadcasting server.