search

LiyingCV / Long-Range-Grouping-Transformer

Implementation of "Long-Range Grouping Transformer for Multi-View 3D Reconstruction" [ICCV 2023]
MIT License
36 stars 1 forks source link

One question on F-Score #12

Open yesterday-Luo opened 5 months ago

yesterday-Luo commented 5 months ago

When I testing on ShapeNet dataset with F-score, I find it will waste too much time. A little help from you would be greatly appreciated!

LiyingCV commented 5 months ago

Hi, we test F-score based on open3d. It will waste time since it needs to sample points. For the computing details, please refer to Legoformer, we use similar methods except for a different open3d version.

yesterday-Luo commented 5 months ago

Thanks for your replying! And I have another question about "Evaluation on the Pix3D Dataset". Could you tell me some training details when you train on ShapeNet Chair dataset with random background from SUN database?

LiyingCV commented 5 months ago

Thanks for your replying! And I have another question about "Evaluation on the Pix3D Dataset". Could you tell me some training details when you train on ShapeNet Chair dataset with random background from SUN database?

Hi, the training configuration is similar to the experiments on ShapeNet except we only train a few epochs on ShapeNet Chair dataset.

eolgaeng-i commented 2 days ago

I’ve been referencing the Pix2Vox issue as well, but I’m still getting different results. Would it be possible for you to kindly share the full code for calculating the F1-score?

Thank you in advance for your help!

LiyingCV commented 2 days ago

I’ve been referencing the Pix2Vox issue as well, but I’m still getting different results. Would it be possible for you to kindly share the full code for calculating the F1-score?

Thank you in advance for your help!

Hi, the code we used is the same as Legoformer. We have tested the F-score of the pre-trained Pix2Vox++, unfortunately, we also obtained the F-Score which is different from Pix2Vox++ paper. Besides, we also find similar wrong results in the issue 76,. Thus we recommend directly using the test code of Legoformer.