Add Semantic Segmentation of KITTI-STEP Panoptic Segmentation Dataset #1748
Conversation
|
|
|
Hi, very thankful for your nice PR. We would review it ASAP. |
Codecov Report
@@ Coverage Diff @@
## master #1748 +/- ##
==========================================
- Coverage 89.04% 88.81% -0.23%
==========================================
Files 144 145 +1
Lines 8636 8674 +38
Branches 1458 1463 +5
==========================================
+ Hits 7690 7704 +14
- Misses 706 730 +24
Partials 240 240
Flags with carried forward coverage won't be shown. Click here to find out more.
Continue to review full report at Codecov.
|
| @@ -0,0 +1,35 @@ | |||
| _base_ = './kittistep.py' | |||
There was a problem hiding this comment.
Sorry, I didn't see any difference between the kittistep.py and kittistep_368x368.py
Would you like to tell us the difference?
There was a problem hiding this comment.
Yes you are right... There is no difference. ...368x368.py could be deleted!
docs/en/dataset_prepare.md
Outdated
|
|
||
| After registration, the data images could be download from [KITTI-STEP](http://www.cvlibs.net/datasets/kitti/eval_step.php) | ||
|
|
||
| You may need to follow the following structure for dataset preparation after downloading iSAID dataset. |
configs/_base_/datasets/kittistep.py
Outdated
| img_dir='training_openmmlab/image_02/train', | ||
| ann_dir='panoptic_maps_openmmlab/train', |
There was a problem hiding this comment.
I have a little question, is KITTI-STEP for panoptic segmentation tasks or semantic segmentation? MMSeg only works on semantic segmentation, and panoptic segmentation and instance segmentation are supported in MMDet
There was a problem hiding this comment.
KITTI-STEP is for panoptic segmentation. But because it follows the Cityscapes label set and the same camera setting (car dashcam view), it is a beautiful dataset to compare sem-seg methods especially if they address ML-interpretatbility. E.g., with this dataset you an compare interclass dependencies or interclass confusion on similar datasets.
There was a problem hiding this comment.
I see. Thanks for your explanation.
Is there any public research work using it as a sem-seg dataset, since the baseline result in MMSeg needs a public reference?
There was a problem hiding this comment.
Thanks for your effort. The closest baseline can be seen here: http://www.cvlibs.net/publications/Weber2021NEURIPSDATA.pdf in the appendix G. They report STQ metric which is based on SQ(=mIoU). So far, there is no published work using it as a pure Sem-seg dataset. But i am working on it so that i hope it will in some months :) If you think it is more convenient, we also could wait some months and I can create a new PR after publication.
There was a problem hiding this comment.
Of course, we really appreciate it that users release their work in mmseg!
There was a problem hiding this comment.
So should i wait with the dataset PR until i used it in a publication?
…#2730) Note that this PR is a modified version of the withdrawn PR #1748 ## Motivation In the last years, panoptic segmentation has become more into the focus in reseach. Weber et al. [[Link]](http://www.cvlibs.net/publications/Weber2021NEURIPSDATA.pdf) have published a quite nice dataset, which is in the same style like Cityscapes, but for KITTI sequences. Since Cityscapes and KITTI-STEP share the same classes and also a comparable domain (dashcam view), interesting investigations, e.g. about relations in the domain e.t.c. can be done. Note that KITTI-STEP provices panoptic segmentation annotations which are out of scope for mmsegmentation. ## Modification Mostly, I added the new dataset and dataset preparation file. To simplify the first usage of the new dataset, I also added configs for the dataset, segformer and deeplabv3plus. ## BC-breaking (Optional) No BC-breaking ## Use cases (Optional) Researchers want to test their new methods, e.g. for interpretable AI in the context of semantic segmentation. They want to show, that their method is reproducible on comparable datasets. Thus, they can compare Cityscapes and KITTI-STEP. --------- Co-authored-by: CSH <40987381+csatsurnh@users.noreply.github.com> Co-authored-by: csatsurnh <cshan1995@126.com> Co-authored-by: 谢昕辰 <xiexinch@outlook.com>
Motivation
In the last years, panoptic segmentation has become more into the focus in reseach. Weber et al. [Link] have published a quite nice dataset, which is in the same style like Cityscapes, but for KITTI sequences. Since Cityscapes and KITTI-STEP share the same classes and also a comparable domain (dashcam view), interesting investigations, e.g. about relations in the domain e.t.c. can be done.
Note that KITTI-STEP provices panoptic segmentation annotations which are out of scope for mmsegmentation.
Modification
Mostly, I added the new dataset and dataset preparation file. To simplify the first usage of the new dataset, I also added configs for the dataset and deeplabv3plus.
BC-breaking (Optional)
No BC-breaking
Use cases (Optional)
Researchers want to test their new methods, e.g. for interpretable AI in the context of semantic segmentation. They want to show, that their method is reproducible on comparable datasets. Thus, they can compare Cityscapes and KITTI-STEP.