search

lkeab / gaussian-grouping

[ECCV'2024] Gaussian Grouping for open-world Anything reconstruction, segmentation and editing.
https://arxiv.org/abs/2312.00732
Apache License 2.0
607 stars 43 forks source link

Could you share the checkpoint and ply files for the trained models used in the paper? #16

Open seokhunchoi opened 8 months ago

seokhunchoi commented 8 months ago

Hi. Thank you for your nice work.

Could you share the checkpoint and ply files for the trained models used in the paper?

I think it would help a lot.

Thanks.

ymq2017 commented 8 months ago

Hi, we will organize all the checkpoints. But it may take some time because there are a lot of datasets and tasks. If you need checkpoints for some individual datasets, I can provide here in advance.

seokhunchoi commented 8 months ago

Checkpoint and ply files trained on LERF-mask dataset (figurines, ramen, teatime) would be a lot of help!

Also, may I know when you will be able to share LERF-mask dataset?

ymq2017 commented 8 months ago

Hi, we update the LERF-mask dataset now and provide a script for calculating IoU and Boundary IoU metric in our paper.

For checkpoints of these three datasets and rendering code with the help of grounded-sam, we will provide them this week.

seokhunchoi commented 8 months ago

Thank you for sharing!

I'm looking forward to the release of checkpoints and the segmentation code using grounded-sam.

I have some questions.

  1. In the lerf-mask dataset, teatime's test views are 0, 1, and 2, but the test masks are only 0 and 2, so there is no test mask for view 1. Is this right?
  2. Also, ramen's test views are 0, 1, 2, and 3, but the test masks are only 0, 1, and 2, so there is no test mask for view 3. Is this right?

Thank you!

ymq2017 commented 8 months ago

Hi, we have updated the checkpoints and segmentation code with text-prompt now.

For the dataset question, yes we omit the annoations on those two views. Because I found these two images from original lerf-loc dataset but I did not find the corresponding camera pose for these two images in their released data. So the baseline result of LERF in Table1 does not include these two images and ours also does not include for fair comparison. And from the rendering result we can find the mask quality is similar on these two images.

seokhunchoi commented 7 months ago

It was very good. Thank you!

Could you also share other checkpoints and ply files such as mipnerf-kitchen, bear, truck etc? For the truck dataset, it would be very helpful to provide the converted truck data with camera parameters you trained with.

Thank you for your kind response.