The Scene type classification task involves classifying a scan into 13 scene types.

Evaluation and metrics

Our evaluation ranks all methods according to recall (TP/(TP+FN)) as well as the PASCAL VOC intersection-over-union metric (IoU = TP/(TP+FP+FN)), where TP, FP, and FN are the numbers of true positive, false positive, and false negative predictions, respectively.



This table lists the benchmark results for the scene type classification scenario.




Method Infoavg iouapartmentbathroombedroom / hotelbookstore / libraryconference roomcopy/mail roomhallwaykitchenlaundry roomliving room / loungemiscofficestorage / basement / garage
sorted bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort bysort by
LAST-PCL-type0.738 10.250 31.000 10.895 11.000 11.000 11.000 10.500 11.000 10.500 20.842 10.000 20.941 10.667 1
Yanmin Wu, Qiankun Gao, Renrui Zhang, and Jian Zhang: Language-Assisted 3D Scene Understanding. arxiv23.12
multi-taskpermissive0.646 20.500 11.000 10.789 20.333 30.667 31.000 10.500 11.000 11.000 10.778 20.000 20.833 20.000 3
Shengyu Huang, Mikhail Usvyatsov, Konrad Schindler: Indoor Scene Recognition in 3D. IROS 2020
3DASPP-SCE0.556 30.500 10.938 30.778 30.667 21.000 10.250 30.500 10.750 30.333 30.500 40.000 20.812 30.200 2
SE-ResNeXt-SSMA0.355 40.000 50.684 40.696 40.200 50.500 40.200 40.500 10.429 40.200 40.545 30.111 10.556 40.000 3
Abhinav Valada, Rohit Mohan, Wolfram Burgard: Self-Supervised Model Adaptation for Multimodal Semantic Segmentation. arXiv
resnet50_scannet0.231 50.200 40.481 50.346 50.250 40.250 50.000 50.500 10.333 50.000 50.357 50.000 20.286 50.000 3