BenchmarksFacade Semantic Segmentation

Facade Semantic Segmentation Benchmark

ZAHA

We observe that albeit deep-learning-driven methods achieve high performance on various tasks, the facade semantic segmentation still poses an unresolved challenge.

See the ZAHA WACV’25 paper introducing the challenge. Would you like to try and solve it? Please grab the data and report your results!

Facade Semantic Segmentation at LoFG3

LoFG stands for Level of Facade Generalization. See more in the ZAHA paper introducing the concept.

PointNetPointNet++Point TransformerDGCNN
OA59.966.475.071.1
P46.137.852.753.6
R42.235.954.745.8
F138.734.852.144.5
IoU26.425.641.633.4
Class scores
wall61.168.576.883.8
window25.626.343.164.1
door13.57.819.821.6
balcony25.10.077.566.7
molding22.543.458.057.5
deco0.00.05.00.0
column22.433.40.037.2
arch19.225.450.22.6
stairs16.00.07.55.6
ground surface12.00.024.421.3
terrain53.553.557.668.0
roof18.76.866.357.4
blinds4.62.318.520.0
interior59.769.172.888.0
other42.747.170.674.1

Exemplary results of the deployed semantic segmentation networks for LoFG3

Facade Semantic Segmentation at LoFG2

LoFG stands for Level of Facade Generalization. See more in the ZAHA paper introducing the concept.

PointNetPointNet++Point TransformerDGCNN
OA71.975.578.282.6
P69.673.075.880.0
R68.173.076.681.8
F168.172.676.180.4
IoU55.859.863.968.5
Class scores
floor92.387.690.792.1
decoration26.247.147.070.0
structural60.965.567.085.2
opening28.227.236.066.2
other el.71.271.678.988.8

Exemplary results of the deployed semantic segmentation networks for LoFG2