New datasets on large scale outdoor point cloud for instance segmentation (HAIS used as baseline, coded are provided)
meidachen opened this issue · 5 comments
Hi @outsidercsy, I really appreciate the amazing work you have done on HIAS, and we did benchmark it on our dataset. I just want to put a note on your repo here in case you or anyone else is interested. We have a new point cloud dataset (STPLS3D, https://github.com/meidachen/STPLS3D) on large-scale outdoor environments which can be used for instance segmentation benchmarking purposes.
Thanks for benchmarking HAIS on STPLS3D! I have updated the information about STPLS3D on our project page. Most point cloud datasets for instance segmentation (ScanNet, S3DIS, SceneNN, etc.) only focus on indoor scenes. Whether the existing methods work if transfered to outdoor scenes is still unkown. We do need a dataset to benchmark their effectiveness in outdoor environments. STPLS3D is a valuable and impressive dataset highly required by the community. Looking forward to your update!
Thank you very much for your encouraging words, and for adding STPLS3D to your homepage!
Hi @outsidercsy , I just want to let you know that we are organizing the Urban3D workshop at ECCV this year with two competitions focusing on semantic and instance segmentation, in case you are interested in participating. I have used HAIS as the baseline again in the instance segmentation track and have uploaded the result to our evaluation server.
workshop: https://urban3dchallenge.github.io/
instance segmentation codalab: https://codalab.lisn.upsaclay.fr/competitions/4646#learn_the_details-overview
semantic segmentation codalab: https://competitions.codalab.org/competitions/31519
Sorry to bother you again, but we have recently identified a critical bug in our evaluation script which affects all of our instance segmentation results. We have fixed the problem and re-evaluated both HAIS and PointGroup. Here are the updated results:
I have updated the result on our STPLS3D repo, could you please kindly update the result shown on your page as well?
Thank you! And sorry for any inconvenience this may be caused.
It's OK. And I have already updated the result.