Problem Formulation:

• Given: a point cloud
• Task: render a RGB image of same scene

Dataset Deliverables

• Subset of the Stadt Wien mobile mapping dataset (531918 7130 x 7130 RGB images (2TB), 1297 .laz point cloud files (131GB) initially only useable by ELISE consortium, planned to be released as Open Government Data (OGD) soon)
• City of Vienna has permission to release the entire dataset as Open Government Data (OGD) (~200 TB raw data) ➡ we decided to also add a Data Set Assembly Tool s.t. researchers can customize the dataset to their needs (eg., using a different resolution, etc.)

PyTorch Dataset and Dataloader

• Dataset and -loader that pre-processes the data, s.t. Elise researchers can readily use the data to develop (generative) models Evaluation Utils & Metrics
• Evaluation Utilities:

1. Fréchet Inception Distance (FID) and Multiscale Structural Similarity Index Measure (MS-SSIM)

2. Intersection over Union (IoU) of pairs of semantic segmented images (real and corresponding fake)

3. Ipython Tutorial Notebook (Explains usage of contained components)

4. Whitepaper