Changing the dataset

[bansilol]

Hi there! I hope you are doing well. I was trying to understand the data acquisition pipeline of the project here. I am a student and for me it is not possible to download the whole kitti dataset at one go(limited data in one day) and hence the download_kitti.py ends up with broken data and I get the error of none type has no attribute shape. Is it possible to somehow train the network on a smaller part of the dataset? I know it might affect the performance of the network but at the end I am doing this as a personal project to learn working with pytorch and depth estimation. If it is possible to change the size of train test val dataset please guide me how can I do it. Many thanks in advance!

Regards, Bansi

[klingner]

Hi there, I think in this case, it is pretty hard to change the sets, because they are pretty fixed defined in the dataloader.

However, if I understand correctly, the only problem is the amount of data downlaoded. I think I would recommend to download and extract the raw data from kitti manually (or in smaller portions). Afterwards, it should be possible to apply the rest of the download_kitti.py script and to run the code as intended.

Regards, Marvin