Semantic segmentation in ISPRS-Potsdam dataset

Semantic segmentation in ISPRS-Potsdam dataset

Semantic segmentation in ISPRS-Potsdam dataset

Authors

  • Maritza Corona Hernandez
  • Pankaj Rajoria

Dataset

The dataset used in our experiments is the 2D Semantic Labeling Contest - Potsdam

The data set contains 38 patches (all images have a dimension of 6000 x 6000 pixels), each consisting of a true orthophoto (TOP) extracted from a larger TOP mosaic. The ground sampling distance of both, the TOP and the DSM, is 5 cm. The DSM was generated via dense image matching with Trimble INPHO 5.6 software and Trimble INPHO OrthoVista was used to generate the TOP mosaic.

Model

The model employed is UNetformer.

image

For further reference: UNETFORMER Architecture

Results

We obtained the following results:

Model Description
Baseline Same as GeoSeg GitHub
ResNet18 We trained with same backbone on GitHub
ResNet50 Using ResNet50 as backbone



Metric   Model  
  ResNet18 ResNet50 Baseline
F1 92.6 92.3 92.6
OA 91.4 90.8 91.2
mIoU 86.5 85.9 86.5


image