A system for videoconferencing that offers, among other features, extremely low end-to-end delay as well as very high scalability. The system accommodates heterogeneous receivers and networks, as well as the best-effort nature of networks such as those based on the Internet Protocol. The system relies on scalable video coding to provide a coded representation of a source video signal at multiple temporal, quality, and spatial resolutions. These resolutions are represented by distinct bitstream components that are created at each end-user encoder. System architecture and processes called SVC Thinning allow the separation of data into data used for prediction in other pictures and data not used for prediction in other pictures. SVC Thinning processes, which can be performed at video conferencing endpoints or at MCUs, can selectively remove or replace with fewer bits the data not used for prediction in other pictures from transmitted bit streams. This separation and selective removal or replacement of data for transmission allows a trade-off between scalability support (i.e. number of decodable video resolutions), error resiliency and coding efficiency.