preprocessing script?

[djamahl99]

I hope this message finds you well. I am currently trying to rerun your code but noticed that there is no implementation for 0_preprocess.py. The script indicates that I should run dufomap, but I am unsure about the specifics of how the dataset was created for training. Could you kindly provide this script or offer some guidance on how to proceed?

Thank you very much for your assistance.

[Kin-Zhang]

We will open it soon at the same time as the paper available in Springer. I will remind you here when it’s public.

Guidance: It uses dufomap and HDBSCAN as said in the paper, I attached both code links for you to have a quick look. We save the label in the h5py file which you can follow this instruction

Last but not least, for people who urgently need to reproduce results, I directly uploaded the pre-trained weight of SeFlow and a mini demo proceeded dataset for a mini-training try.

[djamahl99]

Hi, I can see you have given a link to download the preprocessing on Waymo but AV2 does not seem to have a link?

E.g. waymo you gave:

wget https://zenodo.org/records/12751363/files/waymo_map.tar.gz

Below you have given a hyperlink to preprocessing which is broken (preprocess folder does not exist).

[Kin-Zhang]

Hi, I think you misunderstood the demo data, the one you attached is the preprocessing for Waymo, not Waymo demo data. I didn't upload any Waymo demo data due to their license constraint. The one you screenshot hyperlink it's itself README as I named it different during develop (as dataprocess is transferring raw data to h5py files, process is processing h5py data to have labels inside data for SSL.) You can scroll to learn the command for generating h5py file. I will fix the hyperlink soon.

The demo data link is: https://zenodo.org/records/12751363/files/demo_data.zip?download=1 for av2 Attach readme screenshot for how you can use the link I attached here.

[djamahl99]

Hi, thanks for helping me. I was confused as I thought you were saying you had only omitted preprocessing for nuScenes but indeed you have not provided preprocessing for any datasets (to get labels for SSL). It is necessary that I retrain with your method for my paper so I will have to wait for your preprocessing technique to be released. I am eager to trial your method in future :).

Thanks, Djamahl

[Kin-Zhang]

We will open it soon at the same time as the paper available in Springer. I will remind you here when it’s public.

Guidance: It uses dufomap and HDBSCAN as said in the paper, I attached both code links for you to have a quick look. We save the label in the h5py file which you can follow this instruction

Last but not least, for people who urgently need to reproduce results, I directly uploaded the pre-trained weight of SeFlow and a mini demo proceeded dataset for a mini-training try.

I think I explained here already.

I will remind here when paper is published officially and process SSL label script will also be pushed.