Panoptic Segmentation
For panoptic segmentation, the model output is the combination of instance masks and semantic masks. Ideally, treating each semantic mask as a special “instance” mask.
Embedded Files
0) Overview:
Dataset:
Metrics:
Purpose:
1) Papers:
Paper List:
Panoptic Segmentation Kirillov et al. (Kirillov et al., 2019b) first formulated the task of panoptic segmentation. Top-down methods usually append a semantic segmentation model or branch to the instance segmentation models (e.g., Mask R-CNN) and further fuse the semantic and instance results. Heuristic post-processing steps (Kirillov et al., 2019b, a; de Geus et al., 2018; Li et al., 2019) or specialized fusion modules (Xiong et al., 2019; Lazarow et al., 2020; Ren et al., 2021; Mohan & Valada, 2020) are used to resolve the inherent overlapping problem. Bottom-up methods typically start with semantic segmentation and then cluster ‘thing’ pixels into instances. These methods include SSAP (Gao et al., 2019) and DeepLab-based series works (Yang et al., 2019; Chen et al., 2020a; Wang et al., 2020a, 2021).
References:
Page updated
Google Sites
Report abuse