search

sayands / sgaligner

[ICCV 2023] SGAligner : 3D Scene Alignment with Scene Graphs
https://sayands.github.io/sgaligner/
MIT License
81 stars 4 forks source link

Can you please share an example input? #7

Closed glennliu closed 4 months ago

glennliu commented 4 months ago

Hi

Thanks for your interesting work.

I'm trying to run the code but it seems the data generation is a bit complex. I'm at the step of generating labels.instances.align.annotated.v2.ply using ShunCheng's code. And his repo also requires a few dependencies which are conflict with my Pytorch version.

So, can you provide a pair of processed scene graph inputs to let me quickly run your code? So I can skip the data generation at the moment.

Thanks Chuhao

glennliu commented 4 months ago

After fix quite a few hard-coded data directory, I stopped at running preprocess.py. There is a hard-coded file.

REL2IDX_SCANNET8 = common.name2idx(osp.join(define.SCAN3R_ORIG_DIR, 'files/scannet8_relationships.txt'))

I don't have such file after rungenerate_subscans.py.

sayands commented 4 months ago

Hello,

  1. For now, I have added two example scenes data in the .npy format which is expected by preprocess.py to example_data/ folder. I will also add the code in a while, however, the idea is just porting it in the .npy format after generating labels.instances.align.annotated.v2.ply for faster processing. I cannot provide the dataset because it is ideally derived from 3RScan.

  2. scannet8_relationships.txt contains the 8 relationships predicted by SceneGraphFusion. I have uploaded the file to the pretrained checkpoint GDrive (link in README). However, if you want to use 3RScan GT relationships instead, use the relationships.txt provided by the dataset (just change model:rel_dim parameter to 41 in this case).

Hope this helps, let me know in case of more issues.

glennliu commented 4 months ago

Thanks!

I have generated data.npy and download the scanner_relationship.txt. Will try run evaluation now.

glennliu commented 4 months ago

more file is missing. I asked in a new issue.