No module named 'FFT_MQ'

[swanlu]

Hello, I tried to run your code, but there was a package called "FFT_MQ" that couldn't find. I tried to search it online or in the project of ROSE^1, but I couldn't find any package with the same name. Maybe this is a package designed by yourself?

Hope to get your answer! thank!

[goldleaf3i]

Hi!

Sorry for the issue, I removed from the code pulled here the references to ROSE^1 (so to use the own ROSE^1 repo instead) but I forgot to add the file used for "calling" Rose (which was FFT_MQ). Now it should work.

Let me know if you have any further troubles, so I can try to replicate the error; I used this code on my machine but I didn't have the time to do a fresh install on a new machine for testing.

In a couple of months, I plan to do a small refactoring+documentation of the code, but for now, the code is a bit "rough".

[swanlu]

Thank you very much for your reply and help! Currently FFT_MQ can successfully find the ROSE^1. However, there are still some problems, which I think are caused by the different versions of ROSE^1 we use. Some functions invoked in FFT_MQ are not in my FFTStructureExtraction. Could you help me with this question?

btw I look forward to your further adjustment. This is a groundbreaking room segmentation algorithm.

[goldleaf3i]

Hi, yes, that is probably the issue. We wrote the method and did the experiments more than 1 year ago and before refactoring ROSE. I pushed the version of ROSE I was using online, tomorrow I will check on the laptop I've used to do the experiments for the paper if there are any further differences. Let me know if it works.

Thanks :) As you will see there is still some work to do! The main interest from us was the structure detection, segmentation is more a byproduct. For making segmentation more robust changing the clustering step could be a good idea (it is something that we want to further investigate after we close other stuff).

[tkucner]

Hi, Could you please tell me which parts are you especially interested in FFT_MQ? I have on my machine the core code without relation to room reconstruction.

On Thu, Mar 24, 2022 at 4:13 PM goldleaf3i @.***> wrote:

Hi, yes, that is probably the issue. We wrote the method and did the experiments more than 1 year ago and before refactoring ROSE. I pushed the version of ROSE I was using online, tomorrow I will check on the laptop I've used to do the experiments for the paper if there are any further differences. Let me know if it works.

Thanks :) As you will see there is still some work to do! The main interest from us was the structure detection, segmentation is more a byproduct. For making segmentation more robust changing the clustering step could be a good idea (it is something that we want to further investigate after we close other stuff).

— Reply to this email directly, view it on GitHub https://github.com/goldleaf3i/declutter-reconstruct/issues/1#issuecomment-1077673479, or unsubscribe https://github.com/notifications/unsubscribe-auth/AB37CP773CF3FSMWBPX5PBLVBR2ADANCNFSM5RQICQYQ . You are receiving this because you are subscribed to this thread.Message ID: @.***>

-- Tomasz Piotr Kucner, Ph.D. Post-Doctoral Researcher, Robotics

Mobile Robotics & Olfaction Lab, Centre for Applied Autonomous Sensor Systems, Örebro University,

70182 Örebro, Sweden Room T1105 Phone +46 (0)19 30 14 82 WWW: https://mro.oru.se/people/tomasz-kucner WWW: http://tpkucner.eu Calendar: https://meet-me-7jcc8.appointlet.com/b/tomasz https://meet-me-7jcc8.appointlet.com/b/tomasz

I sometimes opt to work during non-standard hours. I don’t expect that you will read, respond to, or action this message outside of your regular working hours.

[swanlu]

@goldleaf3i Thank you for your help. Now I can successfully run ROSE^2! As you said, I also found some problems with the clustering (especially the thick wall). But I think it will still be a highly robust algorithm for different radar maps. I will keep trying to test your algorithm with the data I collected. Thanks for sharing!

@tkucner Hello! In fact, I'm exploring the effectiveness of different algorithms for room segmentation in indoor robots. Most methods have the problem of insufficient robustness to varying degrees (due to the imprecision of radar mapping)

[goldleaf3i]

@swanlu I am curious about the output if you have any maps to share. We tested the method with the map available to us - 2d lidar-based maps from ground robots. One thing that is important for clustering and structure detection is the scale of the map. If you keep the same scale for all the maps the parametrization for structure detection is rather fixed, but if you have maps with a different resolution it may be that you have still to tune some parameters. But that should not change if your maps are all similar. While clustering for room detection sometimes still gives weird outputs in the case of particular maps (i.e. large-scale ones or if there are walls with several gaps).

Thank you for your interest in our work!