search

QVPR / Patch-NetVLAD

Code for the CVPR2021 paper "Patch-NetVLAD: Multi-Scale Fusion of Locally-Global Descriptors for Place Recognition"
MIT License
517 stars 72 forks source link

Indoor experiment #32

Closed oym1994 closed 2 years ago

oym1994 commented 2 years ago

Hi, thanks for your great contribution. Here I want to consult some questions about indoor experiment.

In the paper, all experements are conducted in outdoor datasets. I wander what would be the difference if applying it into a real indoor enviroment, such as home, official and even some room with repeat region(such as a Server Room where many server computers with the same apperiance are placed)? How can I get the training dataset without GPS in indoor room? Thanks for your attention and I am always looking forward to your kind response and any advice.

Best regards, Slamer

Tobias-Fischer commented 2 years ago

Hi Slamer,

Indoor place recognition brings its own challenges indeed. There are some indoor place rec datasets - see e.g. the VPR-Bench paper for an overview.

You don't necessarily need GPS information. To train the network all you need is a list of query images, a set of potential positives for each query, and a list of definite negatives for each query. GPS information provides a convenient way of obtaining those positives and negatives.

I would recommend reading through some recent papers on indoor place recognition to see how this is typically being done in that context.

Please let us know if you have any more specific questions.

Best, Tobi

oym1994 commented 2 years ago

Hi, Thanks for your kind response. As you have advised, I should read through some papers related to indoor place recognition. Waiting for your next great work!

Best, Slamer