This paper deals with the progressive transmission of large natural scenes over networks, with messages that either drop randomly or arrive with random delays. The unstable networks are called lossy networks.

The authors propose a progressive representation of plants--an n-tree--based on a skeletal representation of the plant. To preserve the branching structure of a plant under low resolution, it is organized hierarchically as a set of generalized cylinders. The proposed scheme jointly clusters similar branches and achieves fine coding efficiency. Additionally, Mondet et al. compare the proposed method with bzip2, an open-source data compressor. The results reveal that the proposed mechanism outperforms the bzip2 compression ratio.

Generally, the paper presents an efficient compression of the plant geometry representation that uses generalized cylinders and is highly suited for data transmission over lossy networks. The authors employ a Bézier curve to reduce the data transmission to represent a branch of a plant, and use a multi-resolution model to encode a plant as a compressed representation. First, the authors group similar Bézier curves. They compute an average curve for each group, and then encode the differences between the control points of a Bézier curve and the average curve, thus saving a huge volume of transmission data. Subsequently, the authors propose the standardization of comparing and coding the differences between two Bézier curves by calculating the average Bézier curve, giving a near representation instance, coding each Bézier curve, and storing the detailed difference vectors. Finally, the authors encode a plant into binary chunks, using four types of data, and transmit the data over networks.

This paper enhances an earlier proposed scheme [1], and gives a suitable, progressive, and dynamic representation for rendering a three-dimensional (3D) scene. The experiments achieve fabulous results in transmission performance.

In summary, the paper provides a detailed derivation and analysis for networked virtual environment applications. It is well structured, with fine methodologies and analyses, and I recommend it.