v0.2.1-beta

[0.2.1-beta]

Fixed

• Fixed bad any_cast in screenshot function

v0.2.0-beta

[0.2.0-beta]

Added

• Added Cubemap Array textures
• Added possibility to create textures from device tensors
• Added function to remove entities from a scene
• Added possibility to remove entities via the scene hierarchy panel
• Added Python bindings for Entity Management in the scene class
• Added Point lights
• Added Shadow mapping for point light on and from meshes (other draw types currently not supported)
• Added function to draw singular components of an entity
• Added possibility to override the shaders used by an entitie's component
• Added possibility to override the shaders of RendererSystem::draw on meshes
• Added entity.entity_handle() method
• Added Performance Panel to monitor frame time
• Added multi sampled textures
• Added multi sampled framebuffers
• Added multi sample anti aliasing (MSAA)
• Added context manager for graphics contexts that enables more robust multi-threaded rendering
• Added dedicated shader compiler class
• atcg::CameraController::onUpdate now returns if the camera was updated in the current frame
• Added revision counter to Transform components
• Added copy methods to Graph and Camera
• Added RevisionStack which allows for Ctrl+Z and Ctrl+Y
• Added simple include functionality for shaders
• Added shader subroutines
• Added ATCG_DEBUG for debug messages
• Added explicit models for Camera extrinsics and intrinsics
• Added functions to convert between opengl and opencv intrinsics
• Added width and height to camera component
• Added camera preview in frustum rendering
• Added possibility to load camera images to be rendered
• Added main scene camera
• Added fly-to and set from view camera buttons
• Added scripting engine and python scripting to entities with a Script Component
• Added slider to control the size of rendered camera frustums
• Added PixelUnpackBuffers
• Added possibility to set texture data from PixelUnpackBuffers
• Added options to configure the python side build
• Added draw method to scene
• Added skyboxes to scene
• Added Dictionaries
• Added Skybox class
• Added a more flexible render pass system that allows also the use of predefined renderpass
• Added a component renderer class that handles rendering of different components

Changed

• The Renderer now internally also uses the Texture ID push system
• Cube maps are no longer interpreted as depth 6 textures
• Entity lookup in a scene by name is now approximately O(1) via hashing
• If an Entity can't be found by ID, an invalid entity is returned
• Circle shader now also writes out entity IDs
• Removed RendererSystem::draw function that takes in a material struct
• At the end of each frame, all texture units are unbinded to create a clean state for the next frame
• It is now possible to attach arbitrary textures (2D, 3D, Cube, etc.) to framebuffers. This changed the return value of fbo->getColorAttachament() from atcg::ref_ptr to atcg::ref_ptr
• fbo->blit now explicitely tries to copy all color attachements
• RendererSystem::useScreenBuffer now uses the MSAA framebuffer if MSAA is enabled (default). If MSAA is turned off, the old framebuffer and behavior is used
• RendererSystem::getFramebuffer now returns the defacto state of the last frame. Direct render calls to this framebuffer will have no errect if MSAA is enabled because it will get overwritten by the blitting of the framebuffers. Use RendererSystem::getFramebufferMSAA() instead. This behavior is also the case for getFrame and getZBuffer
• Graphics Contexts can no longer be created by hand but have to be created via the ContextManager namespace.
• atcg::CameraController::onUpdate now returns if the camera was updated in the current frame
• drawGuizmo now takes in the scene as an additional parameter
• operator bool() of an entity now also checks if the entity handle is still part of the scene
• IDComponent::id is no longer read-write but read only
• NameComponent::name is no longer read-write but read only
• glm objects can no longer be directly casted to numpy arrays but have to be converted explicitely using .numpy()
• Removed unnecessary wrapper class around logger, now everything directly works on an spdlog object
• Constructor of PerspectiveCamera now uses CameraIntrinsics and CameraExtrinsics
• Camera Controllers can no longer be initiated with an implict camera, but it has to be created and passed manually
• Torch DLLs are NO LONGER COPIED to the output directory on Windows. The user has to make sure the PATH variables point to the correct DLLs.
• The PythonFrontend.py no longer compiles the project. It now requires that the package is installed via pip.
• The python folder was moved into atcg_lib
• Reworked design of RenderGraph
• The atcg::RendererSystem is now no longer responsible to render scenes and handle cubemaps. This is now done in the scene class
• Renderer::getFramebuffer now returns the internally used framebuffer which might be a MSAA framebuffer if MSAA is enabled. Specific getters for both framebuffers were added

Fixed

• Screenshot functionality of Renderer uses correct member method and not global renderer instance.
• scene->removeAllEntities now clear internal buffers
• Fixed Texture tests on CPU backend
• Fix normals of cylinder in test scene
• Normals of cylinder in test scene
• Improved code of the Renderer and reduced code duplication
• Fixed hanging texture references when a texture is deleted
• Fixed managing of framebuffer state in multi threaded applications
• Fixed crash when setting faces of a graph with a device tensor (because torch can't sort GPU uint32). Current workaround copies the data to CPU first
• Fixed bindings for glm objects in python
• Fixed GUI for camera extrinsics in SceneHierarchyPanel

v0.1.1-beta

This patch fixes a minor missing include in the system registry. Otherwise, it updates all subpackages and adds some of them as explicit submodules.

[0.1.1-beta]

Added

• Updated external libraries

Fixed

• Added missing include in system registry

v0.1.0-beta

This update adds

• TCP Server and Clients
• Changed the structure of the core systems that the framework uses (Renderer, ShaderManager, etc.) that makes it possible to have multiple instances of these classes which can be used for multi-threaded rendering etc.
• Adds headless rendering (on linux even without hidden window states, so no X-server is needed)
• Worker Threads and Thread Pools
• Command Line arguments
• Fullscreen and other more advanced window options
• More unit tests and more robustness in general by adding asserts

The full changelog:

[0.1.0-beta]

Added

• Possibility to flip decoded jpeg images
• Custom key codes
• Methods to read and write structs from network buffers
• Add python bindings for network utils
• atcg::shader_directory() and atcg::resource_directory() to get absolute paths to shader and resources
• ATCG_TARGET_DIR for each target that is built to get information about it's location
• A path for atcg::ShaderManagerSystem can now be specified
• Default imgui layout
• Add Worker and WorkerPool for parallalized CPU computing
• The framework can be installed via pip
• Possiblity to index individual layers of a TextureArray
• create method for Cubemaps from tensor data
• Added support for unsigend data types in torch tensors
• Applications can now be set to fullscreen via F11 or via window->toggleFullscreen()
• Add Rendergraph and Renderpasses to implement more complicated rendering algorithms
• Add unit tests for shader and shader manager
• Add function to set window position
• Add functions to change the decoration states of windows
• Add atcg::getCommandLineArguments() to access command line arguments

Changed

• Default behavior of JPEG encoder and decoder: now flip images
• The caller of TCPClient::sendAndWait had to manually prepend the message size. This is now done automatically.
• The callback functions of TCPServer now expect std::vector references instead of raw pointers.
• Indices are now consistently handles as uint32_t, also for tensors.
• Each Renderer now has it's own instance of a shader manager.
• Each Renderer now has control over it's own context.
• The documentation now uses a nicer theme and is separated into multiple sections.

Fixed

• Fixed linking of imgui in headless mode by removing unnecessary libraries in linking stage
• Fixed crash when using the JPEG modules without initializing the torch allocator
• Channel size of RG texture format
• isHDR() function for RG textures
• GPU memcopy of float 3D textures
• GPU memcopy of Texture arrays
• Mark non-compute shader as such if a compute shader gets recompiled