Weighted boxes fusion (WBF) is an object detection ensemble method that averages bounding box predictions from multiple models rather than discarding redundant boxes. It works by clustering overlapping boxes, calculating a weighted average bounding box for each cluster, and rescaling confidence scores based on cluster size. This allows all predictions to contribute to the final output. WBF is shown to outperform non-maximum suppression (NMS) and soft-NMS on ensembles of different object detection models and with test time augmentation, producing more accurate averaged predictions.