search

valeoai / RADIal

162 stars 52 forks source link

Raw Data Preprocessing #25

Closed illaMcbender closed 2 years ago

illaMcbender commented 2 years ago

Hi, thank you for sharing the code! It is a really great job! I've been following your work recently and found that you've provided raw dataset and some preprocessing code in SignalProcessing folder. But I failed to preprocess the raw data into the "ready to use" format that you provided in https://drive.google.com/file/d/1PpAcL5r2PRYMxDb46ASps9YqToL7uJcE/view I' m wondering if you can provide the complete code about how to process the raw data into the "ready to use" dataset ? Thanks a lot!

jrebut commented 2 years ago

Hi,

Please, find here the script to generate the dataset from the raw data. The script was design at the beginning of the project and the code has been clean-up to be shared. SO, I m not sure it is still compatible with the current signal processing library. ANyway, this is pretty simple, it just saves images, laser data and radar per frame basis Please, have a look and let me know if you need further assistance

Julien

illaMcbender commented 2 years ago

Thank you so much for your assistance! I will have a try!

JeroenBax commented 2 years ago

Hi, Thanks for the helpfull script, I was wondering which radar processing library you used in this script and if this is still downloadable.

jrebut commented 2 years ago

Hi, I dropped all the scripts here. I will try in the following weeks to add some instructions as this is a recurrent request.

message shared with another user:

Basically, the dataset has something like 30k frames and even more. Not everything was released. The labelling process was semi automatic. That is to say, object detection algorithm were run on both laser data and camera. If these 2 algorithm agreed, it generated a label candidate. Finally, label candidates were reviewed, corrected or discarded. This is why there is much less frames with labels than the full dataset size. So, the label candidates file is, as far as I can remember, a selection of key frames with their objects position. This first selection provides positive frames. Now, with the same script, there is a second label candidates file for negative frame, with no object. I will provide it to you.

To summarize: 1- run the script with label candidates for positive frames 2- run the script with label candidates for negative frames 3- both should give the ready to use dataset

The script was designed like 2 years ago with a cython version (named rpl) of a c++ code to execute fft, cfar and AoA. This was replaced by the python signal processing library available here

sohrab-m commented 11 months ago

Hi @jrebut , links in the previous comments do not seem to work anymore. I am wondering if you could provide an updated link?