Limited Reconstructions Semantic Instance Results

The 3D semantic instance prediction task involves detecting and segmenting the object in an 3D scan mesh.

Our evaluation ranks all methods according to the average precision for each class. We report the mean average precision AP at overlap 0.25 (AP 25%), overlap 0.5 (AP 50%), and over overlaps in the range [0.5:0.95:0.05] (AP). Note that multiple predictions of the same ground truth instance are penalized as false positives.

This table lists the benchmark results for the 3D semantic instance with limited reconstructions scenario.

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

InstTeacher3D	0.473 1	0.778 3	0.635 1	0.331 4	0.322 2	0.732 1	0.089 1	0.502 1	0.200 1	0.350 1	0.440 1	0.507 1	0.489 1	0.493 3	0.327 1	0.655 1	0.469 1	0.924 1	0.268 2

WS3D_LR_Ins	0.440 2	0.704 4	0.533 2	0.485 1	0.335 1	0.659 2	0.025 5	0.460 2	0.156 2	0.315 2	0.412 2	0.254 2	0.383 2	0.638 2	0.278 2	0.653 2	0.431 2	0.869 2	0.330 1
Kangcheng Liu: WS3D: Weakly Supervised 3D Scene Segmentation with Region-Level Boundary Awareness and Instance Discrimination. European Conference on Computer Vision (ECCV), 2022
TWIST+CSC	0.342 3	0.796 2	0.412 6	0.233 6	0.170 3	0.637 3	0.028 2	0.292 4	0.089 3	0.214 3	0.320 3	0.133 3	0.337 4	0.444 4	0.242 3	0.458 5	0.328 6	0.852 3	0.179 3
Ruihang Chu: TWIST: Two-Way Inter-label Self-Training for Semi-supervised 3D Instance Segmentation. CVPR 2022
CSC_LR_INS	0.322 4	0.824 1	0.433 4	0.288 5	0.169 4	0.610 5	0.028 2	0.302 3	0.084 4	0.174 5	0.274 4	0.061 5	0.370 3	0.358 5	0.142 5	0.507 4	0.367 3	0.701 6	0.097 6

PointContrast_LR_INS	0.304 5	0.233 6	0.418 5	0.389 2	0.136 6	0.596 6	0.006 6	0.274 5	0.038 6	0.140 6	0.265 5	0.052 6	0.314 5	0.647 1	0.149 4	0.533 3	0.350 5	0.806 4	0.132 4

Scratch_LR_INS	0.287 6	0.485 5	0.438 3	0.362 3	0.154 5	0.614 4	0.028 2	0.218 6	0.058 5	0.191 4	0.247 6	0.065 4	0.272 6	0.271 6	0.109 6	0.433 6	0.355 4	0.766 5	0.102 5

3D Semantic Instance with Limited Reconstructions Benchmark