**DeferredGS: Decoupled and Editable Gaussian Splatting with Deferred Shading** This paper introduces DeferredGS, a method for decoupling and editing the Gaussian splatting representation using deferred shading. Gaussian splatting is a rendering technique that approximates 3D scenes with a set of 3D Gaussians, offering real-time performance but lacking editing capabilities. DeferredGS addresses this by modeling illumination with a learnable environment map and defining additional attributes such as texture parameters and normal directions on Gaussians. The normal directions are distilled from a signed distance function (SDF) to enhance geometry reconstruction. Deferred shading is applied to enable more realistic relighting effects compared to previous methods. Experiments demonstrate that DeferredGS produces superior results in novel view synthesis and editing tasks, including faithful decomposition and editing of geometry, texture, and lighting. The method is evaluated on synthetic and real datasets, showing improved performance over baseline methods in terms of novel view synthesis, decomposition, and relighting.**DeferredGS: Decoupled and Editable Gaussian Splatting with Deferred Shading** This paper introduces DeferredGS, a method for decoupling and editing the Gaussian splatting representation using deferred shading. Gaussian splatting is a rendering technique that approximates 3D scenes with a set of 3D Gaussians, offering real-time performance but lacking editing capabilities. DeferredGS addresses this by modeling illumination with a learnable environment map and defining additional attributes such as texture parameters and normal directions on Gaussians. The normal directions are distilled from a signed distance function (SDF) to enhance geometry reconstruction. Deferred shading is applied to enable more realistic relighting effects compared to previous methods. Experiments demonstrate that DeferredGS produces superior results in novel view synthesis and editing tasks, including faithful decomposition and editing of geometry, texture, and lighting. The method is evaluated on synthetic and real datasets, showing improved performance over baseline methods in terms of novel view synthesis, decomposition, and relighting.