Actually the length encoding described by StefanB is not Varint32, but part of the Syntactic Description Language used to define Expandable classes as documented in ISO_IEC_14496-1-System. The idea is that the definition of an object can contain its own length fields to account for future new objects while still allowing legacy code to read the current definitions. If a parser sees a tag id it doesn't understand, it can skip over it using the encoded length data immediately succeeding it.