How davwed0/senior-design-project works
Main Components
• Pipeline Orchestrator: Represent the main controller (run_pipeline.sh) that initiates the entire process.
• Input Datasets: Include a component that depicts datasets (with folders for <cluster_name> and <base_cluster>, along with files like mask.pt, PNG/JPG images, cameras.bin, images.bin, points3D.bin, database.db).
• Environment & Dependencies: Note the Linux environment, Conda environments (“r3dg” and “sugar”), and a CUDA-compatible GPU.
• R3DG Submodule: Show the Relightable3DGaussian submodule, which carries out two rounds of training (initial training and second training).
• NEILF Training: Illustrate this as a step between R3DG training runs to add relighting support.
• Output Generation: Include steps/components for Mesh and Texture Extraction and the final outputs (showing separate folders under Relightable3DGaussian/output and SuGaR/output).
Detail the Relationships and Interactions
• How the pipeline first passes data to R3DG for initial training.
• Direct the flow from NEILF training into the SuGaR component, emphasizing the refinement interaction.
• Finally, demonstrate paths leading to the second round of R3DG training and NEILF training to integrate the refined results, and then onward to Mesh and Texture Extraction.
• Finally, point out that all these components ultimately produce output files stored in the output/ hierarchy, mapping both early-phase (point clouds, json configurations) and final mesh outputs.
Architectural Patterns
• Modularity: The use of submodules (Relightable3DGaussian and SuGaR) suggests a modular design where each component can be developed, tested, or replaced independently.
• Sequential/Pipelined Processing: Indicate that the pipeline has clear sequential stages (initial training, refinement, re-training, and extraction) with feedback loops.
Incorporate Technologies & Conventions
• Mention the use of shell scripts (run_pipeline.sh) as the orchestrator.
• That the pipeline leverages CUDA for GPU support.
• Reference the external repositories (cloned as submodules) and their roles in the overall processing.
Specific Elements
• Labeled Blocks/Nodes: Represent each major module or processing stage.
• Environment Boundaries: Optionally add a note or a symbol to indicate the different Conda environments and the GPU dependency.