Dive 3.2 Keypoints

• Pointset - Pointsets are a unified way of describing multi-polygons, triangular and quadratic meshes, polylines and set of vertices. Pointsets contain a table of vertices, which are indexed in order to build the polygons and such. This reduces file size, memory occupation and network traffic. Pointsets are also an easy way of performing polygon deformations, such as those needed for body animation.
• New Dive Node types - New node types have been added:
• Switch nodes decide which of their children should be used for rendering, interactions, collisions, etc. Switch nodes are an easy way of creating simple animations from Dive/Tcl scripts or C programs.
• Billboards rotate around an axis or their center to always look at a user. Billboards are an easy way of creating textured trees, birds, etc.
• Level of Details have been improved and are now a special node type. Level of details can choose which representation to use depending on the distance or the angle between the viewer and the object.
• Texture transforms - Texture coordinate rotation and translation matrices have been added to allow control of texture mapping via transform operations. This makes it possible to rotate or translate textures in different ways on the mapping surface.
• New VRML 1.0 parser - The parser is not anymore based on a BNF grammar, but on a dynamic grammar. Such a grammar is necessary since VRML 1.0 files can be self-describing. The new parser optimize the number of objects, views and material created and is adapted to the new pointsets. Special VRML nodes exist for specific Dive information such as properties and scripts.
• New vishnu interface - The interface of vishnu, Dive's default viewer has been improved. It contains a Netscape-like button bar which can be detached from the main window. Several new functionalities have been added:
• On-the-fly edition of all the materials associated to an object is now possible using a material editor.
• New simplification rendering flags have been added for low-end platforms.
• All top windows can be raised through a new menu.
• Support for subjective views and AC3D (see below) is present.
• Several entity (actor, object, world, etc.) information windows can be opened simultaneously.
• All notifications since the beginning of a session are accessible from an history window.
• The entity inspector is enhanced and an "inspect by name" tool has been added.
• Positions and Orientations as user defined functions - The position and orientation of objects within the virtual world can be described by a set of associated functions. Even if some distribution problems remain, the following user-defined functions are offered:
• angular and linear velocities associated to objects are a special case of these general functions.
• gravity and collision detection have been implemented.
• Dive/Tcl commands can be called each time the position and orientation of an object is requested. This allow easy implementation of complex mathematical based movements and constraints.
• Faster state transfers - The content of state transfers can be compressed, which results in faster state transfers, even on a local network. This improves the dynamic of connection and disconnection of processes, since senders are not blocking as long as before.
• RTP v2 - All Dive packets are now encoding according to the IETF standard for real-time communication. Dive sessions can now coexist on the same multicast address as RTP v2 compliant tools, such as VIC, VAT or RAT.
• Text-to-speech - An experimental and simple text-to-speech interface has been added to all platforms supporting audio. Sound generated from the text-to-speech synthesis will locally be spatialized and mixed with all Dive sound. As an example, messages received in the "talk" windows can be synthesized as human voice. The voice will be spatialized as coming from the sender's representation.
• Audio Encoding - GSM and DVI encoding have been added to the platform, in order to improve long distance audio communication and to reduce band-width use.
• Video Support - Dive processes can send and receive video streams on a separate channel. Streams can dynamically bound to textures. As a result, only powerful texture machines can be used to visualize video streams. The frame rate is a function of CPU power (for decompression) and texture binding time. Encoding is only possible on SGI workstations.
• The DiveBone - Dive proxyservers can now be connected to each others in order to connect together islands of multicast networks. A transparent mechanism ensure that the traffic from regular Dive processes on a local island will be forwarded to all other islands, and thus be received by possible remote Dive processes. Proxyservers can be controlled by a separate Tk-based windowing interface, which will be able to display their current status and some network statistics. This setup allows remote control of the different proxyservers running along the DiveBone.
• Network analysis - Real-time and delayed network analysis is possible using the following tools:
• All incoming and outgoing Dive events can be logged to a file, together with most of the content.
• Proxyservers can regularily send statistics to their controlling Tk interface. The statistics include the number of bytes sent and received for each application entry, i.e. for each event type.
• Subjective views - Subjective views are a way of slightly tweaking how users see the virtual environment, e.g. making some objects transparent or brighter for some users. Edition and use of subjective views is possible from new menu entries in vishnu.
• AC3D - AC3D is a modeller supporting different file formats, including the Dive file format. The following key points are now part of Dive 3.2:
• Dive can export entity trees to the new AC3D file format.
• During a vishnu session, objects can be exported to a running AC3D, in which edition is possible. AC3D objects can then be put in place of the original version.

Emmanuel Frécon - emmanuel@sics.se
Olov Ståhl - olovs@sics.se