This is Pytorch re-implementation of our CVPR 2020 paper "Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation" (https://arxiv.org/abs/1911.10194)
Apache License 2.0
585
stars
117
forks
source link
[Discussion] Changing Instance Segmentation Head to 2D Detection Head #112
Hi there (@bowenc0221, @tingliu, @PkuRainBow , @YknZhu , @mcollinswisc),
Do you think that changing the instance segmentation head to a detection head like YOLO is feasible/a good idea? Seems like the instance seg head and object detection head are very alike. For my application, I do not really need panoptic segmentation but semantic segmentation + object detection will be more suitable.
I have read some panoptic methods actually does object detection first and then post-merge with semantic segmentation to form panoptic segmentation, did you perform any experiment in this area for your network design? I am very curious to know your thoughts on this to gain additional insight before I proceed to re-design the network.
Hi there (@bowenc0221, @tingliu, @PkuRainBow , @YknZhu , @mcollinswisc),
Do you think that changing the instance segmentation head to a detection head like YOLO is feasible/a good idea? Seems like the instance seg head and object detection head are very alike. For my application, I do not really need panoptic segmentation but semantic segmentation + object detection will be more suitable.
I have read some panoptic methods actually does object detection first and then post-merge with semantic segmentation to form panoptic segmentation, did you perform any experiment in this area for your network design? I am very curious to know your thoughts on this to gain additional insight before I proceed to re-design the network.