Jiangnan gardens, a prominent style of Chinese classical gardens, hold great potential as digital assets for film and game production and digital tourism. However, manual modeling of Jiangnan gardens heavily relies on expert experience for layout design and asset creation, making the process time-consuming. To address this gap, we propose GardenDesigner, a novel framework that encodes aesthetic principles for Jiangnan garden construction and integrates a chain of agents based on procedural modeling. The water-centric terrain and explorative pathway rules are applied by terrain distribution and road generation agents. Selection and spatial layout of garden assets follow the aesthetic and cultural constraints. Consequently, we propose asset selection and layout optimization agents to select and arrange objects for each area in the garden. Additionally, we introduce GardenVerse for Jiangnan garden construction, including expert-annotated garden knowledge to enhance the asset arrangement process. To enable interaction and editing, we develop an interactive interface and tools in Unity, in which non-expert users can construct Jiangnan gardens via text input within one minute. Experiments and human evaluations demonstrate that GardenDesigner can generate diverse and aesthetically pleasing Jiangnan gardens.
Overview of the GardenDesigner pipeline. GardenDesigner transforms the user input into a Jiangnan garden through Hierarchical Garden Composition and Knowledge-Embedded Asset Arrangement. First, Hierarchical Garden Composition transfers the user input into parameters for terrain and road generation with aesthetic principles. Subsequently, Knowledge-Embedded Asset Arrangement resolves garden asset selection and relational constraint-based layout optimization for each area: (1) Asset Selection chooses the objects based on the garden knowledge and area information; (2) Layout Optimization assigns the object according to garden knowledge, and then optimization loss is used to get the feasible solution for the garden layout.
GardenVerse construction from Internet repositories and manual modeling. We invite experts to modify the architectures and construct object combinations. Finally, garden experts annotate the basic information and garden knowledge for assets.
We evaluate generated Gardens from physical plausibility, structure complexity, semantic coherence, and aesthetic quality. 1. Pathway Score (Path-S). Path-S is used to determine whether significant plants and buildings can be reached or viewed along the road. 2. Class Diversity (Class-Div). Additionally, we also use Class Diversity to measure the object categories' diversity. 3. Fractal Dimension (FD). We calculate structure complexity. 4. CLIP-Score is used to measure the consistency between the generated garden and the instruction. And the garden type prompt template is in the form of ''a top-down view of Jiangnan garden scene feature.'' 5. CLIP-Aesthetic is used to evaluate the aesthetic score. 6. VLM-Score. We prompt the VLMs to rate the rendered garden image. 7.QA-Quality. We also used VLM-based visual scorer Q-Align to evaluate results.
This paper has proposed GardenDesigner, a novel framework that encodes aesthetic principles for Jiangnan garden construction and integrates a chain of agents procedurally. By structuring the generation process into hierarchical garden composition and knowledge-embedded asset arrangement, GardenDesigner ensures spatial rationality and aesthetic coherence with Jiangnan garden design principles. This integration not only facilitates the creation of authentic garden layouts but also serves as a platform for preserving and exploring classical Chinese landscape architecture in digital contexts. Looking forward, GardenDesigner can be extended to support interactive educational tools, virtual heritage reconstruction, and personalized landscape design, opening new avenues for cultural heritage preservation and creative applications in digital art and games.
