Jungle Ruins
This blog describes the scene-level technical considerations needed to overcome real-time path tracing difficulties. Jungle Ruins combines high-frequency detail, enormous instancing, and alpha testing into one workload.
Jungle Ruins is a research-driven scene that explores several open challenges in real-time path tracing. It was created as a large-scale natural environment scene inspired by the beauty and complexity of the Amazon rainforest. Intel Labs’ Visual Compute and Graphics lab researchers Cristiano Siqueira, Manu Mathew Thomas, Miroslaw Pawlowski, and Anton Sochenov developed the project. Technical Artist Cristiano Siqueira creates datasets and model workloads for real-time rendering.
The scene is envisioned as a dense, lifelike rainforest environment, serving as a sandbox for integrating and stress-testing research while providing a visually rich scene for creative exploration. It supports complex workflows, rendering optimisation, and large virtual environments.
One of the key goals of Jungle Ruins is to consolidate multiple technical challenges into a single, unified workload. This approach allows the researchers to expose and analyze complex interactions between challenges that might not be apparent when addressed individually. The core challenges unified in this workload include:
- High-frequency Detail: Rendering fine textures or sharp edges in complex scenes with dense geometry is difficult. Using only one sample per pixel leads to aliasing and exponentially increases the complexity for denoising techniques.
- Massive Instancing: Essential for realistic forests and dense vegetation, this involves often millions of instances. This scale introduces significant challenges for memory and acceleration structure management in real-time path tracing. Each unique geometry requires separate Bottom-Level Acceleration Structures (BLAS), increasing memory consumption and build times. Top-Level Acceleration Structures (TLAS) grow, affecting traversal efficiency and performance.
- Alpha Testing: Evaluating opacity maps at every ray intersection for vegetation is computationally demanding in path tracing. This significantly increases memory access for high-resolution textures and ray traversal time through multiple semi-transparent layers, putting considerable strain on performance in scenes with dense foliage. An alternative, using fully modeled leaf geometry, avoids alpha testing overhead but can dramatically increase triangle counts, potentially exceeding technical budgets and GPU memory constraints.
Creating Jungle Ruins involved intricate planning and creative problem-solving, following a workflow that included:
- Conceptualization and Research: Identifying rendering features and techniques to showcase, translating technical details into creative ideas, and drawing inspiration from the Amazon rainforest using nature documentaries, photographs, and Artificial Intelligence -generated concept art.
- Asset Building: Balancing artistic ambition with technical constraints using a modular design and procedural generation approach.
- Terrain: The foundation is an 8Kx8K procedural landscape, approximately 64km on a real-world scale. To manage this vast area efficiently, the terrain was divided into 64 individual 1km tile sets. This allowed for precise control over scattering and improved viewport responsiveness. Subdividing the high-detail core region allowed finer modifications and faster scattering optimisations.
- The vegetation was classified as High (canopy trees), Medium (medium-sized trees), and Low (seedlings, bushes, groundcover). A modular design concept mechanically generated and reused atomic components like leaves and branches across plant groups for visual consistency and reduced manual burden. Procedural node trees with seed parameters helped streamline the creation of countless variations quickly.
- Bringing It All Together: Using Blender as the hub application, baseline assets were imported, and the scene was composited. The tiled terrain allowed for precise control over foliage scattering using full-resolution geometry for scattering to maintain natural complexity.
Scalability and modularity were central to the technical design, supporting research goals. The tiling method allowed the artist to focus on asset creation while researchers could efficiently load and experiment with individual tiles, enabling rapid prototyping and evaluation of rendering techniques without slowing content creation. This modular setup simplified targeted adjustments, iterative testing, and streamlined complexity management and performance optimizations.
Optimizing the scene for the target platform, which uses low-end GPUs, presented challenges due to strict technical budget requirements. The research framework also supported only basic scene features. To address this, asset optimization was prioritized, involving simplifying base geometry, using a limited number of materials, and strategically distributing texture resolutions. A significant portion of the limited unique triangle budget was concentrated on the surface vegetation as it is most visible when instanced. Rendering techniques were explored to boost performance, such as disabling alpha testing where its visual impact was minimal.
Skipping alpha tests for geometry beyond a certain distance from the ray origin reduced unnecessary visibility checks. This technique could improve total rendering time by up to 10%. Specific issues, like color shifting with the pyramid’s moss when alpha testing was triggered, were resolved by recreating the base geometry to closely follow the opacity texture outline, eliminating the need for an opacity map. Framework limitations, such as basic support for PBR materials and direct texture inputs, prevented the use of tiled textures for terrain, resulting in less detail in close-up views in certain areas like cliffs. Despite constraints, shifting the triangle budget toward high-quality vegetation and carefully scattering foliage helped enhance the scene’s appeal and conceal terrain imperfections.
The project highlights the benefits of real-time path tracing for lighting and rendering pipelines, fundamentally simplifying workflows for creators. Key benefits demonstrated include:
- Accurate Dynamic Lighting: Path tracing eliminates the need for pre-baked lighting. Global illumination, shadows, and reflections are computed dynamically, allowing artists to immediately see accurate lighting updates when moving lights or adjusting materials, resulting in a much faster “what you see is what you get” workflow.
- Simplified Asset Authoring: Assets can be authored more straightforwardly using physically accurate properties. The path tracer simulates light behavior under a unified physics model, handling various materials consistently. Artists no longer need special case shader tricks or workarounds like baked ambient occlusion. Assets built with true PBR values look correct because the algorithm inherently captures lighting effects.
- Consistency Between Authoring and Final Render: Path tracing greatly narrows the gap between content previewed during creation and its appearance in the final render. Lighting and shading adhere to physical correctness at runtime, bringing results much closer to the ground truth. This means fewer surprises and less tweaking late in development, accelerating convergence toward the artistic vision.
Jungle Ruins is described as an ever-evolving series pushing the boundaries of content creation and consumption. Building the scene provides a solid platform for content creation and study, advancing path-tracing applications in production situations. As part of the commitment to open software, the scene is licensed under Creative Commons, allowing researchers and artists to build on it.