MACnet: Mask augmented counting network for class-agnostic counting

•Class agnostic counting usually uses bounding boxes to identify the target exemplar object that will be counted.•We argue that bounding boxes are not ideal for class agnostic counting since it captures both foreground and background.•We propose Mask Augmented Counting Network (MACnet) that integrates segmentation mask to better capture foreground features.•We improve upon state-of-the-art by up to 3.7 MAE points on the FSC-147 benchmark dataset.•Integrating segmentation mask is nontrivial. We design a residual module that learns to combine mask and exemplar features.