2D Semantic Instance Results - ScanNet Benchmark

The 2D semantic instance prediction task involves detecting and segmenting the object in an image.

Our evaluation ranks all methods according to the average precision for each class. We report the mean average precision AP (from overlaps [0.5:0.95:0.05]), as well as AP 50% for an overlap value of 50. Note that multiple predictions of the same ground truth instance are penalized as false positives.

Method	avg ap	bathtub	bed	bookshelf	cabinet	chair	counter	curtain	desk	door	otherfurniture	picture	refrigerator	shower curtain	sink	sofa	table	toilet	window

UniDet_RVC	0.205 2	0.381 2	0.323 3	0.037 3	0.226 3	0.177 3	0.063 2	0.277 3	0.120 1	0.067 3	0.131 3	0.074 3	0.317 2	0.080 3	0.235 1	0.289 3	0.141 3	0.678 1	0.080 3

EMSANet (Instance)	0.241 1	0.401 1	0.439 1	0.085 1	0.242 1	0.220 1	0.081 1	0.289 2	0.117 2	0.121 1	0.182 1	0.126 1	0.346 1	0.181 2	0.181 2	0.358 1	0.156 1	0.675 2	0.131 1
Seichter, Daniel and Fischedick, Söhnke and Köhler, Mona and Gross, Horst-Michael: EMSANet: Efficient Multi-Task RGB-D Scene Analysis for Indoor Environments. IJCNN 2022
FKNet	0.204 3	0.334 3	0.358 2	0.038 2	0.234 2	0.184 2	0.025 3	0.318 1	0.042 4	0.088 2	0.141 2	0.053 4	0.300 3	0.207 1	0.171 3	0.292 2	0.149 2	0.636 3	0.109 2

MaskRCNN_ScanNet	0.119 4	0.129 4	0.212 4	0.002 4	0.112 4	0.148 4	0.014 4	0.205 4	0.044 3	0.066 4	0.078 4	0.095 2	0.142 4	0.030 4	0.128 4	0.139 4	0.080 4	0.459 4	0.057 4
Re-implementation of Kaiming He, Georgia Gkioxari, Piotr Dollár, Ross Girshick: Mask R-CNN. ICCV'17

2D Semantic Instance Benchmark