Question about generated semantic maps

[ZhengLeon]

Hi @srama2512,
Thanks for your great work.
After generating the semantic maps like this:
1-LXt-Fkjw3q-L-1.png
I can't successfully calculate the fmm_dists.
May I ask why there are no semantic annotations in my semantic maps? Thanks a lot.

[srama2512]

@ZhengLeon - thanks for your interest in our work. What procedure did you use to generate the map? Can you provide the steps to reproduce?

[ZhengLeon]

@srama2512 , thanks for your kind reply.
I used the codes as follow to convert the raw MP3D scenes to GLB:

data_path = "/mnt/PONI/data/scene_datasets/mp3d_uncompressed/v1/scans"
first_directory = os.listdir(data_path)

for second_directory in first_directory:
  mesh_directory = os.path.join(data_path, second_directory, 'matterport_mesh')
  scene_directory = os.listdir(mesh_directory)
  for scene in scene_directory:
    obj_file_path = os.path.join(mesh_directory, scene, scene+'.obj')
    glb_file_path = os.path.join(mesh_directory, scene, second_directory+'.glb')
    if os.path.exists(obj_file_path):
      print(obj_file_path)
      scene = trimesh.load(obj_file_path)
      scene.export(glb_file_path)

Then, I use the below script to extract MP3D semantic maps:
ACTIVE_DATASET="mp3d" python scripts/create_semantic_maps.py

And the generated GT_info is attached here:
semmap_GT_info.json.zip

Thanks.

[srama2512]

@ZhengLeon - After converting, did you ensure that the semantic ply files are also present in the scene directory? This is my current file structure

> cd <PATH TO MATTERPORT DIR>/17DRP5sb8fy/
> ls
17DRP5sb8fy.glb  17DRP5sb8fy.house  17DRP5sb8fy.navmesh  17DRP5sb8fy_semantic.ply

[ZhengLeon]

Hi, @srama2512
Yes, there exists semantic ply files. But the .navmesh and .house files are missed. May I ask if they are initiated in the original dataset? Thanks!

[srama2512]

Hi @ZhengLeon,

You can get the full dataset by following the instructions from here. Note that the .glb files you get through this download will be basis compressed. You have to use instead the .glb files generated from .obj.

[ZhengLeon]

@srama2512 , thanks for your guidance.
Now I can successfully generate the semantic maps.
However, I get only 108 train / 21 val maps from precompute_fmm_dists.py, which is different from the paper "153 train / 21 val".
Is something wrong with my split?

[srama2512]

Great! No, there's nothing wrong. The 153 number in the paper is a typo since it includes train + test. Please see this issue comment for more details.

[ZhengLeon]

@srama2512, thanks for helping.
I will close this issue now.