StructAdapt is the companion open-source project to the paper Static Conversion and Deployment of AdaConv for Real-Time 3D Style Transfer in Unity.
This repository provides the engineering implementation of the method described in the paper: exporting the AdaConv style-transfer algorithm to the ONNX format and deploying it inside the Unity Engine. The project addresses the structural incompatibility between AdaConv’s dynamic kernel generation and static-graph inference frameworks by introducing a grouped-convolution vectorization strategy. Through this static conversion approach, three model variants are exported from the same AdaConv checkpoint:
- Fully Dynamic Model – preserves both spatial and batch dimension dynamics.
- Batch-Dynamic Model – fixes spatial dimensions while keeping batch dynamics.
- Static Model – uses a completely static computation graph.
All variants achieve real-time rendering of about 14.5 fps on an RTX 4060 and reach SSIM ≈ 0.67–0.68, demonstrating mathematical equivalence to the original dynamic formulation. This repository therefore serves as a practical blueprint for deploying structurally dynamic neural networks in real-time 3D style-transfer applications that require static inference graphs.
📄 For detailed methodology and experimental results, please refer to the published paper.
