Table 4.

Per-class IoU on the ScanNet Benchmark. The results for DDNet and other state-of-the-art methods were evaluated on the official website of the ScanNet Benchmark Challenge.

Method	PSPNet [28]	Mseg [48]	3DMV [49]	AdapNet++ [50]	FuseNet [11]	DDNet
bathtub	0.490	0.505	0.481	0.613	0.570	0.898
bed	0.581	0.709	0.612	0.722	0.681	0.660
bookshelf	0.289	0.092	0.579	0.418	0.182	0.185
cabinet	0.507	0.427	0.456	0.358	0.512	0.431
chair	0.067	0.241	0.343	0.337	0.290	0.304
counter	0.379	0.411	0.384	0.370	0.431	0.256
curtain	0.610	0.654	0.623	0.479	0.659	0.659
desk	0.417	0.385	0.525	0.443	0.504	0.461
door	0.435	0.457	0.381	0.368	0.495	0.480
floor	0.822	0.861	0.845	0.907	0.903	0.904
other furniture	0.278	0.053	0.254	0.207	0.308	0.293
picture	0.267	0.279	0.264	0.213	0.428	0.572
refrigerator	0.503	0.503	0.557	0.464	0.523	0.484
shower curtain	0.228	0.481	0.182	0.525	0.365	0.785
sink	0.616	0.645	0.581	0.618	0.676	0.748
sofa	0.533	0.626	0.598	0.657	0.621	0.623
table	0.375	0.365	0.429	0.450	0.470	0.284
toilet	0.820	0.748	0.760	0.788	0.762	0.847
wall	0.729	0.725	0.661	0.721	0.779	0.773
window	0.560	0.529	0.446	0.408	0.541	0.548
mIoU	0.475	0.485	0.498	0.503	0.535	0.560