search

serizba / salad

Optimal Transport Aggregation for Visual Place Recognition
GNU General Public License v3.0
140 stars 13 forks source link

Question about dataset orgatization #3

Closed whu-lyh closed 6 months ago

whu-lyh commented 7 months ago

Wao. Amazing job! A new strategy for VPR. Thanks for your releasing codes. I wander how to process the multi-datasets and get the .npy file under dataset dir? Any instructions?

serizba commented 7 months ago

Hi @whu-lyh

Do you mean how to create .npy files for a new dataset, or how to use existing .npy files?

whu-lyh commented 7 months ago

about how to create .npy files

serizba commented 7 months ago

These files are created manually. For SALAD, we just copied the .npy files from GSV-Cities. If you need to create new ones, you can just save the required info for the new dataset (image paths, indexes, ground truths..).

I suggest loading the .npy files of an existing dataset to take a look to the desired labels to save.

whu-lyh commented 7 months ago

Thanks

whu-lyh commented 6 months ago

I found that the MSLS val dataset in your .npy file is different to the official released image data? The number of images is different. Did you do some data clean operation? Thanks.

serizba commented 6 months ago

As said, these files are copied from GSV-Cities. You can take a look here for an explanation on the number of images.

whu-lyh commented 6 months ago

Oops, thanks.