Semantic Segmentation Refinement by Monte Carlo Region Growing of High Confidence Detections

The semantic segmentation produced by most state-of-theart methods does not show satisfactory adherence to object boundaries. Methods such as fully-connected conditional random fields (CRFs) can significantly refine segmentation predictions. However, they rely on supervised parameter optimization that depends upon specific datasets and predictor modules. We propose an unsupervised method for semantic segmentation refinement that takes as input the confidence scores generated by a segmentation network and re-labels pixels with low confidence levels. More specifically, a region growing mechanism aggregates these pixels to neighboring areas with high confidence scores and similar appearance. To minimize the impact of high-confidence prediction errors, our algorithm performs multiple growing steps by Monte Carlo sampling initial seeds in high-confidence regions. Our method provides both running time and segmentation improvements comparable to state-of-theart refinement approaches for semantic segmentation, as demonstrated by evaluations on multiple publicly available benchmark datasets.