Bridging Global Context Interactions for High-Fidelity Pluralistic Image Completion

We introduce PICFormer, a novel framework for Pluralistic Image Completion using a transFormer based architecture, that achieves both high quality and diversity at a much faster inference speed.

Our key contribution is to introduce a code-shared codebook learning using a restrictive CNN on small and non-overlapping receptive fields (RFs) for the local visible token representation.

This results in a compact yet expressive discrete representation, facilitating efficient modeling of global visible context relations by the transformer. Unlike the prevailing autoregressive approaches, we proposed to sample all tokens simultaneously, leading to more than 100× faster inference speed. To enhance appearance consistency between visible and generated regions, we further propose a novel attention-aware layer (AAL), designed to better exploit distantly related high-frequency features. Through extensive experiments, we demonstrate that the efficiently learns semantically-rich discrete codes, resulting in significantly improved image quality. Moreover, our diverse image completion framework surpasses state-of-the-art methods on multiple image completion datasets.

The overall pipeline of PICFormer. (a) It first learns a quantizer using a code-shared strategy, along with a restrictive CNN. (b) A transformer is then applied to infer the composition of the original embedded indices. (c) Finally, we sample the top K results, merge them with the original high-resolution image, and pass them to a refinement network with an Attention-Aware Layer (AAL) to transfer high-quality information from both visible and generated regions. Note that only the bottom pipeline is used during inference, while the top pipeline is for learning the quantizer offline.