Regarding inference for Rx and Geometric Features

[ishaanmht]

Dear Authors,

Thanks a lot for making your valuable work public. Could you provide some guidance on running inference to get some of the other results in your paper like Received power at receiver and Geometric layout of the environment ?
In your experience how accurate is solving the inverse problem for example recovering the layout ?

[tomsal]

Hi,
thanks for your interest in our work! The prediction of the received power is evaluated using the eval_regression.py as described in the readme file. To run the recovery of the geometric layout you will need to employ Wi-GATr as a diffusion model which was not published in this repo. The backbone model remains the same, but it will require additional training to capture the joint distribution of the data. Hope this clarifies it! :)

[ishaanmht]

Hello,

Thanks a lot for clarification. I just had one more question: For the regression task it seems like the dataloader loads only 1 rx point per environment. For diffusion model do you guys use single tx rx pair in training and inference as well ?