Gf-Gssr: Gradient Fused Gaussian Splatting Super-Resolution

In the field of Novel View Synthesis, 3D Gaussian Splatting (3DGS) has shown great potential. However, image resolution is often limited in practice due to factors such as camera limitations, making it difficult to accurately synthesize high-resolution novel views. To address this, we propose Gradient-Fused Gaussian Splatting SuperResolution. By integrating a single-image super-resolution (SISR) strategy, prior knowledge is introduced to adapt the Gaussian field for high-resolution rendering. Additionally, we introduce a gradient fusion module to mitigate errors from the SISR model, ensuring accurate guidance of the 3DGS. Experimental results on the Synthetic NeRF Dataset demonstrate significant improvements, validating the effectiveness of our approach in enhancing high-resolution novel view synthesis.