search

valeoai / carrada_dataset

GNU General Public License v3.0
85 stars 22 forks source link

mis-alignments of camera images and radar views #8

Open Harry-Zhi opened 2 years ago

Harry-Zhi commented 2 years ago

Hi.

As mentioend in the paper, the camera images and radar data are synchronised. However, I found that in the dataset the number of camera images are ususally not the same to that of radar data.

Could you share more information about how to get the corresponding camera image for each radar view?

Many thanks.

ArthurOuaknine commented 2 years ago

Hello.

Could you please provide an example? As far as I remember, the dataset should contain the same number of radar frames and camera images for each sequence.

Harry-Zhi commented 2 years ago

Hello.

Could you please provide an example? As far as I remember, the dataset should contain the same number of radar frames and camera images for each sequence.

Hi. It seems like all the sequences in downloaded carradar dataset have a different number of images and radar views.

For example, in the 2019-09-16-12-52-12 sequence, there are 1015 images in folder "camera images" while 1017 files for RD, RA and AD views.

In 2019-09-16-12-55-51, there are 631 images while 627 radar views.

This happens to all other sequences as well.

Could you have a check at the dataset?

BTW, I used the processed version (~100G after unzip) to report the number above.

ArthurOuaknine commented 2 years ago

I am sorry but I will not be able to access the dataset for several weeks so I won't be able to check the sequence manually. As far as I remember, the radar and camera frames ares well synchronized. However, due to the synchronization, additional fames have been kept, either for the camera or the radar. Thus, I suggest you to trust the indexing of the sensors frames and to discard the additional ones.

I hope it will help. Please let me know if it is not enough and if you observe a synchronization problem.