Jiacheng Chen

> > Uh, sorry for the confusion. The file utils.py was directly copied from FloorNet's codebase... > > FloorNet provides the functions for computing recall/precision for walls(edges) and rooms, these...

Thanks for your interest! Could you provide more details about how the masks are different from the ones in the paper? Maybe uploading some pictures of your generated masks and...

Hi, thanks for your interest. I didn't try this but I think you can use a renderer to get a simple 3d indoor model: just assign a height for every...

Hi, If you can get the RGB-D scans of those complex buildings as well as some corresponding annotation data, you can re-train the whole system and get the results you...

@ogencoglu I have put two links in the readme, and the second one is still working ([http://realsee.com/open](http://realsee.com/open)). It's in Chinese but you can use a translator and read the form.

Hi, Thanks for your interest. You can refer to the preprocessing part in my other repo [here](https://github.com/woodfrog/heat/blob/master/s3d_preprocess/DataProcessing/FloorRW.py#L62C67-L62C83). These preprocessing utils for Structured3D data were inherited from a previous paper, MonteFloor,...

In the current implementation, the number of edges after edge filtering is set to be (3 * N) instead of O(N^2), where N is the number of corners (check [here](https://github.com/woodfrog/heat/blob/master/train.py#L102)...

> Hello, @woodfrog , still a problem about this dataset, our dataset is about 120 images, I augment 10 times to get about 1000 inputs, but the training is still...

The problem might come from your use of nuScenes-mini dataset -- I used the full nuScenes dataset and exported MapTR's outputs as the initial proposals. If you run with nuScenes-mini,...

Indeed, the original page of the dataset seems to be offline now, probably due to some changes made by Beike. Since we do not have the right to distribute Beike's...