Instance and Panoptic segmentation

[abhigoku10]

@LiWentomng thanks for thsi toolbox it shall be very helpfull to the community, just had few queries

can we use this to train instance segmentation using BDD100k dataset for all the scenes
Panpotic based segmentation dataset
Thanks in adavance

[LI Wentong]

@abhigoku10
For BDD100k, the methods in this toolbox can work for instance classes , like car . For the lane and road, I have not try these objects. I guess it may work not well. You can try boxinst or boxlevelset in this toolbox to test the performance first. Any further questions can be discuessed.

For panpotic segmentaiton, the methods, like boxinst, boxlevelset and DiscoBox, can not support this task. Because these methods are based on the instance segmentation framework, CondInst and SOLOv2. The method of Box2Mask-T in this paper, which is based on the transformer-based framework, to achieve the box-supervised task. It can achieve the panpotic segmentaiton (The stuff categories also need the bounding box annotations).
In the left-bottom of this figure, the pizza(thing) and ding table(stuff) can be well segmented. The code of Box2Mask-T is under the preparation, I will release it later.