Multi-task Fusion for Efficient Panoptic-Part Segmentation
This work addresses efficient multi-task segmentation for computer vision applications, offering incremental improvements in a domain-specific area.
The paper tackles panoptic-part segmentation by introducing a network that generates semantic, instance, and part segmentation with a shared encoder and a parameter-free joint fusion module, achieving state-of-the-art results with improvements of up to 4.7 percentage points in PartPQ on datasets like Cityscapes Panoptic Parts.
In this paper, we introduce a novel network that generates semantic, instance, and part segmentation using a shared encoder and effectively fuses them to achieve panoptic-part segmentation. Unifying these three segmentation problems allows for mutually improved and consistent representation learning. To fuse the predictions of all three heads efficiently, we introduce a parameter-free joint fusion module that dynamically balances the logits and fuses them to create panoptic-part segmentation. Our method is evaluated on the Cityscapes Panoptic Parts (CPP) and Pascal Panoptic Parts (PPP) datasets. For CPP, the PartPQ of our proposed model with joint fusion surpasses the previous state-of-the-art by 1.6 and 4.7 percentage points for all areas and segments with parts, respectively. On PPP, our joint fusion outperforms a model using the previous top-down merging strategy by 3.3 percentage points in PartPQ and 10.5 percentage points in PartPQ for partitionable classes.