UA - Gaussian Depth Disc (GDD dataset)

Dear reader, Welcome! You must be an avid profilometry person to be interested in downloading our dataset. Before you start tinkering with the dataset package please do install the requirements.txt libraries for a more easy step into operating this system. We hope to have made the hierarchy of the package as clear as possible! Also note that this system was written in VScode. The Matlab script that creates the surface has been included for if you yourself would like to print one out and do in depth research! Find your way to the examples folder, there you can find "entire_dataset". This folder contains a script to divide the original h5 file (480x640) containing all data into whatever sub-options you'd like (had to be made int16 as to not exceed to 50GB limit of Zenodo, you are more than welcome to reach out for the 32int format!). We included a downsampled format of 120x160 (int32) as this makes the file size more managable for pc's with not enough RAM. An example divided dataset has already been given namely the 80/20 division of respectively training and validation data in the "example_dataset" folder, of the downsampled dataset. In the folder models you will find the two models mentioned in the publication related to this dataset. These two were published with the dataset since they had either the highest performance on the training and validation dataset (DenseNet) or on the random physical object test (CNN). A training script is included (training_script.py) to show you how these models were created, so if you wish to add new models to the networks.py file in the classes folder, you can! The validation jupyter notebook contains two visualisation tools to quickly and neatly show the performance of your model on the recorded dataset. Lastly to test on the recorded objects you can run the "test_physical_data.py" and "test_physical_data_sine_deivation.py" scripts. We hope this helps you in your research and we hope it further improves any and all research within the single shot profilometry field! 😊 Kind regards, Rhys Evans, InViLab, University of Antwerp, Belgium