IAE: Implicit Autoencoder for Point Cloud Self-supervised Representation Learning

Autoencoding has been a popular topic across many fields and recently emerged in the 3D domain. However, many 3D representations (e.g., point clouds) are discrete samples of the underlying continuous 3D surface which makes them different from other data modalities. This process inevitably introduces sampling variations on the underlying 3D shapes. In learning 3D representation, a desirable goal is to disregard such sampling variations while focusing on capturing transferable knowledge of the underlying 3D shape. This aim poses a grand challenge to existing representation learning paradigms. For example, the standard autoencoding paradigm forces the encoder to capture such sampling variations as the decoder has to reconstruct the original point cloud. In this paper, we introduce the Implicit Autoencoder(IAE). This simple yet effective method addresses this challenge by replacing the point cloud decoder with an implicit decoder. The implicit decoder can output a continuous representation that is shared among different point cloud samplings of the same model. Reconstructing under the implicit representation can prioritize that the encoder discards sampling variations, introducing appropriate inductive bias to learn more generalizable feature representations. We validate this claim via experimental analysis. Moreover, our implicit decoder offers excellent flexibility in designing suitable implicit representations for different tasks. We demonstrate the usefulness of IAE across various self-supervised learning tasks for both 3D objects and 3D scenes. Experimental results show that IAE consistently outperforms the state-of-the-art in each task.