Closed ogencoglu closed 11 months ago
You can adapt the current sampling codes to get an inference pipeline for a single image input -- simply cleaning the data loading codes while keeping the necessary pre-processing steps for the input density image. However, to make the model work well for 2D floorplan images instead of density images, you will need to prepare such a dataset and re-train the model.
For an example sample of the density map, there are many in the paper, like the one below:
Or if you download the processed dataset with the link in README, these images are already included.
Thanks a lot. Appreciated!
Thanks for the great work!
In your paper it is mentioned that point clouds are converted to top-view point-density images as inputs.
I was wondering if your work and implementation can be utilized to detect the rooms and walls directly from a real 2D floor plan image? Would you be able to provide a pipeline for a single image input without having to request access and download whole datasets? Or maybe an example sample for the density-map.