FAST-Splat 是一種新穎的語義高斯樣條函數方法,它通過將封閉集語義分割擴展到開放詞彙設置,實現了快速且無歧義的 3D 語義對象定位,同時顯著減少了訓練和渲染時間以及 GPU 內存使用量。
FAST-Splat is a novel method for semantic Gaussian Splatting that achieves faster training and rendering speeds than existing methods while also resolving semantic ambiguity in object localization by directly embedding semantic codes into Gaussian primitives and leveraging a hash table for disambiguation.