Running paper experiments + running Matterport based experiments about neural-slam HOT 3 CLOSED

Training and evaluation instructions are available here:
https://github.com/devendrachaplot/Neural-SLAM#usage

The pretrained models should give you numbers slightly above those reported in the paper.

Regarding Matterport, can you check if the Matterport dataset is working with the habitat-api/habitat-lab (without using the Neural-SLAM code)?

from neural-slam.

Is there a simple set of instructions available for running and reproducing experiments from the paper? I can't seem to find the relevant documentation anywhere.

Also, are there a set of concrete exact steps for running the code on Matterport? I have tried following along the steps provided in #16 but was hoping there was a better way?
Simply using the mp3d configuration file leads to a Trade::AbstractImporter::image2D(): index out of range error.

Thanks!

Hi,
I also got the same error when runing with the mp3d dataset. Did you solve it?

from neural-slam.

I did! I switched the habitat_api folder inside env/habitat (https://github.com/devendrachaplot/Neural-SLAM/tree/master/env/habitat) to the latest version of habitat.
I than transferred the changes from this commit: devendrachaplot/habitat-api@8ac1b16#diff-c0fd02c963c59f63d0125bf919c5fd5cbfeb0c20d8a12a8f7a2a17a5122f1de8 to the new habitat environment.

from neural-slam.