Datasets > Modality > Images > TRN (Toulouse Road Network)

TRN (Toulouse Road Network)

Introduced by Belli et al. in Image-Conditioned Graph Generation for Road Network Extraction

The Toulouse Road Network dataset describes patches of road maps from the city of Toulouse, represented both as spatial graphs G = (A, X) and as grayscale segmentation images.

The TRN dataset contains 111,034 data points (map tiles), of which: 80,357 are in the training set (around 72.4%), 11,679 are in the validation set (around 10.5%), 18,998 are in the test set (around 17.1%).

Each tile represents a squared region of side 0.001 degrees of latitude and longitude on the map, which corresponds to a square of around 110 meters. The semantic segmentation of each patch is represented as a 64 × 64 grayscale image.

The dataset is generated starting from publicly available data from OpenStreetMap. More details on the dataset characteristic and generation methods are available in our blogpost.




  • Unknown