KP-FCNN

About

Category Paris-Lille-3D
Short name KP-FCNN
Long name KPConv Segmentation Network for Point Clouds
URL https://github.com/HuguesTHOMAS/KPConv
Description H. Thomas, J.E. Deschaud, B. Marcotegui, F. Goulette, Y. Le Gall. Semantic Classification of 3D Point Clouds with Multiscale Spherical Neighborhoods. In The IEEE International Conference on Computer Vision (ICCV), 2019.
BibTeX @InProceedings{Thomas_2019_ICCV, author = {Thomas, Hugues and Qi, Charles R. and Deschaud, Jean-Emmanuel and Marcotegui, Beatriz and Goulette, Francois and Guibas, Leonidas J.}, title = {KPConv: Flexible and Deformable Convolution for Point Clouds}, booktitle = {The IEEE International Conference on Computer Vision (ICCV)}, month = {October}, year = {2019} }
Hardware Intel(R) Core(TM) i7-3770 CPU @ 3.40GHz, 32GB RAM, Nvidia GeForce GTX 1080 Ti
Runtime 1000
Used additional training data No
Source code Yes
Created at 22/01/2019 13:03
Updated at 23/01/2020 14:40

Result

Av. IoU Ground Building Pole Bollard Trash_can Barrier Pedestrian Car Natural
Result 82.0 99.5 94.0 71.3 83.1 78.7 47.7 78.2 94.4 91.4
Rank 5 6 18 7 7 2 14 4 7 10

Full confusion matrix

L1' L2' L3' L4' L5' L6' L7' L8' L9'
L1 18082931 23981 275 465 989 5202 121 1192 2924
L2 31248 6694481 216 0 150 8218 7 663 6279
L3 511 14216 119632 126 336 725 919 665 18869
L4 350 81 301 14019 0 56 0 0 108
L5 250 86 0 13 31445 1043 0 489 534
L6 11704 122446 214 1210 4269 162504 52 6 7741
L7 252 3789 0 0 0 63 21660 2 788
L8 2563 43483 0 0 0 1531 55 899232 2771
L9 3363 175329 10769 139 337 13799 0 4 2577904