Abstract: In this paper, we extend the MPU implicits algorithm to deal with unoriented point sets while preserving its desirable properties, such as sharp feature preservation. An orientation inference algorithm is introduced to orientate the local implicit patches by solving a graph labeling problem through energy minimization. Sign consistency between local functions is exploited to infer the globally consistent orientation. To precisely model the features, we employ the affinity propagation clustering algorithm to identify the local surface patches composing the features by considering orientation consistency between data points. Sharp features can then be accurately reconstructed by performing piecewise smooth surface fitting. Experimental results are shown to demonstrate the performance of the proposed algorithm.