Unsupervised Foreground Extraction via Deep Region Competition
This addresses the problem of extracting foreground objects without supervision for computer vision applications, representing an incremental improvement over prior methods.
The paper tackles unsupervised foreground extraction from images by proposing Deep Region Competition (DRC), which reconciles energy-based priors with generative modeling using Mixture of Experts and learned pixel re-assignment, resulting in competitive performance on complex real-world data and multi-object scenes, with potential generalization to unseen object categories.
We present Deep Region Competition (DRC), an algorithm designed to extract foreground objects from images in a fully unsupervised manner. Foreground extraction can be viewed as a special case of generic image segmentation that focuses on identifying and disentangling objects from the background. In this work, we rethink the foreground extraction by reconciling energy-based prior with generative image modeling in the form of Mixture of Experts (MoE), where we further introduce the learned pixel re-assignment as the essential inductive bias to capture the regularities of background regions. With this modeling, the foreground-background partition can be naturally found through Expectation-Maximization (EM). We show that the proposed method effectively exploits the interaction between the mixture components during the partitioning process, which closely connects to region competition, a seminal approach for generic image segmentation. Experiments demonstrate that DRC exhibits more competitive performances on complex real-world data and challenging multi-object scenes compared with prior methods. Moreover, we show empirically that DRC can potentially generalize to novel foreground objects even from categories unseen during training.