about dataset splits

[HongGu-Jeong]

hello!! thank you for your awesome work

I have a question about dataset splits which mentioned in your paper

("We use the train/validation/test set split proposed
by [33], which yields 40 000, 80 and 700 samples for train,
validation and test sets")

i checked [33] ref. paper and find "'city', '2011_09_26_drive_0005_sync' " is validation set, but it have 152 samples( after preprocess ).

This is different from what was mentioned.

can you explain about these differences.

thanks for reading this !

[pclucas14]

Hi,

I don't exacty recall how I did the splits, but from the paper is seems that I subsampled the recordings

For training we subsample from 10 Hz to 3 Hz since temporally adjacent frames are nearly identical.

So maybe this explains why there are less frames. Are you looking to rebuild the dataset I used for the paper ?

[HongGu-Jeong]

Thank you for the quick reply to my question

Yes, I'm looking to rebuild the dataset you used for the paper

And following your previous answers in "Issues", i did it based on /kitti_data/parse_velo.py

I obtained train/validation/test set that is 40814 / 152 / 830 samples respectively.
(after preprocess)

I can't use "processed-kitti-lidar repo" because it does not contain intensity that I want to know

but I checked that the number of [city-2011_09_26_drive_0005_sync](validation set) in both "processed-kitti-lidar repo" and reconstruction dataset which i rebuild are same.

I am a little concerned because the reconstructed dataset and the dataset mentioned in the paper are different.

[pclucas14]

Hi,

So I checked the original training files and 40814 --> 13608 / 152 / 830 looks ok. However, I have never used the intensity value, so the dataset does not have them. I think I only downsampled the training set but not the validation or test sets.

I am not sure what would be the best way to reconstruct the dataset with the intensity values. You would have to redownload the raw files (as done here) and apply my preprocessing script to the files. It will be a pain though; I remember that for some point clouds it didn't work, and I had to manually remove them

Best of luck!
-Lucas

[HongGu-Jeong]

Thank you for the information!