Issues with evaluation of Truck scene in TNT

[yanivw12]

Hello,

I am trying to evaluate on your provided TNT dataset (TNT_GOF) and I'm having some issues with the Truck scene. See the following issue I opened in the 2DGS repository (I used their model and evaluation code, but since the dataset originated from here, it's still relevant):

https://github.com/hbb1/2d-gaussian-splatting/issues/111

The problem seems to be that the Truck_COLMAP_SfM.log file supplied in the TNT_GOF dataset does not match the Truck_trans.txt alignment, resulting in the transformed mesh not aligning with the GT mesh, and therefore achieving a low score:

Can you please confirm that the SfM file you use to achieve the paper results for Truck is the same as the one you uploaded to TNT_GOF? Other scenes work well, the only issue is with the Truck scene.

Also, I'd appreciate if you could share the preprocessing code that generated those files given the original images.

Thanks in advance.

[niujinshuchong]

Hi, it works for me. I think you need to check if you download the gt alignment file correctly.

[yanivw12]

I'm pretty sure that I did - is this the alignment file you used?

[niujinshuchong]

Yes. I think you can run our code to check.

[MagicTZ]

I've got the same issue, do you fix it?

Thanks in advance.

Hello,

I am trying to evaluate on your provided TNT dataset (TNT_GOF) and I'm having some issues with the Truck scene. See the following issue I opened in the 2DGS repository (I used their model and evaluation code, but since the dataset originated from here, it's still relevant):

hbb1/2d-gaussian-splatting#111

The problem seems to be that the Truck_COLMAP_SfM.log file supplied in the TNT_GOF dataset does not match the Truck_trans.txt alignment, resulting in the transformed mesh not aligning with the GT mesh, and therefore achieving a low score:

Can you please confirm that the SfM file you use to achieve the paper results for Truck is the same as the one you uploaded to TNT_GOF? Other scenes work well, the only issue is with the Truck scene.

Also, I'd appreciate if you could share the preprocessing code that generated those files given the original images.

Thanks in advance.

[yanivw12]

I've got the same issue, do you fix it?

Thanks in advance.

Hello, I am trying to evaluate on your provided TNT dataset (TNT_GOF) and I'm having some issues with the Truck scene. See the following issue I opened in the 2DGS repository (I used their model and evaluation code, but since the dataset originated from here, it's still relevant): hbb1/2d-gaussian-splatting#111 The problem seems to be that the Truck_COLMAP_SfM.log file supplied in the TNT_GOF dataset does not match the Truck_trans.txt alignment, resulting in the transformed mesh not aligning with the GT mesh, and therefore achieving a low score: Can you please confirm that the SfM file you use to achieve the paper results for Truck is the same as the one you uploaded to TNT_GOF? Other scenes work well, the only issue is with the Truck scene. Also, I'd appreciate if you could share the preprocessing code that generated those files given the original images. Thanks in advance.

Yes, see my comment here:

https://github.com/hbb1/2d-gaussian-splatting/issues/111#issuecomment-2464840622

[MagicTZ]

I've got the same issue, do you fix it? Thanks in advance.

Hello, I am trying to evaluate on your provided TNT dataset (TNT_GOF) and I'm having some issues with the Truck scene. See the following issue I opened in the 2DGS repository (I used their model and evaluation code, but since the dataset originated from here, it's still relevant): hbb1/2d-gaussian-splatting#111 The problem seems to be that the Truck_COLMAP_SfM.log file supplied in the TNT_GOF dataset does not match the Truck_trans.txt alignment, resulting in the transformed mesh not aligning with the GT mesh, and therefore achieving a low score: Can you please confirm that the SfM file you use to achieve the paper results for Truck is the same as the one you uploaded to TNT_GOF? Other scenes work well, the only issue is with the Truck scene. Also, I'd appreciate if you could share the preprocessing code that generated those files given the original images. Thanks in advance.

Yes, see my comment here:

hbb1/2d-gaussian-splatting#111 (comment)

Thanks a lot, it works!!