search

drprojects / DeepViewAgg

[CVPR'22 Best Paper Finalist] Official PyTorch implementation of the method presented in "Learning Multi-View Aggregation In the Wild for Large-Scale 3D Semantic Segmentation"
Other
222 stars 24 forks source link

Should the kitti360 dataset be complete now, or only a portion of it need to be downloaded #22

Closed myself0816 closed 1 year ago

myself0816 commented 1 year ago

Should the kitti360 dataset be complete now, or only a portion of it need to be downloaded

drprojects commented 1 year ago

Hi, thanks for your interest in this project.

As mentioned in issue #9, you can use a small version of the KITTI-360 dataset by setting mini=True in your dataset config.

More specifically, the 3D data only comes as a single file for KITTI-360, so all the points will be downloaded. But I assume this is not really where the issue is for you, since this file is relatively small. The image files are the heavy ones. Those can only be downloaded sequence by sequence, and camera by camera see the official download page and the script I created to automate downloading.

I just updated the code to only download the images of the sequence you need, which, if you set mini=True, should allow you to work with sequence 0 for the train and val datasets and sequence 8 for the test set. Please let me know if this works for you. 😉

Obviously enough, if you eventually want to train a multimodal model on KITTI-360, you will need to download the images for all sequences, which I agree is a bit much but I have no control over, since the dataset was built that way.

myself0816 commented 1 year ago

Thank you for your answer. Since I can't connect to Amazon, can you share how the kitti360 dataset was downloaded?

drprojects commented 1 year ago

The KITTI-360 dataset is indeed hosted using Amazon AWS, I invite you to refer to the official dataset website. As per how I coded the dataset auto-download, please check out the script mentioned above.