The rendering result is completely black, and the problem is located

[scott198510]

I have identified the issue with the rendering result.
The current version of the code, when pulled directly, produces a completely black effect.

After my comparison, I found that the main part of the code does not need to be changed.
What needs to be changed is the diff -gaussian-rasterization in the submodules .
Specifically, in the current branch 9c5c202, the rendered result has depth information, so this version's result is completely black, which is not suitable.

If you switch the submodules/diff Gaussian rasterization to the main branch, the rendered result is correct.
Unfortunately, only the depth rendering result is missing.

If you are interested, you can run a comparison to find the final bug in the submodules.

My understanding is that the current version of the code does not match the branch 9c5c202 of diff-gaussian-rasterization marked inside, resulting in almost no significant decrease in loss during rendering.

Either the main body of the code needs to be changed to adapt to the branch 9c5c202 of diff-gaussian-rasterization , or the main body of the code needs not be changed and the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information.

@yzslab @jakubcerveny @emepetres @gdrett

[ENGFCC]

Hello, my friend. I meet the same problem as you, did you solve it? Could you please tell me how to make it? I can't know what to do about "the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information". Especially, what is "9c5c202"? I guess it might be a link, but I can't click it.

[lcicl]

Hello, my friend. I meet the same problem as your, did you solve it? Could you please tell me how to make it? I can't know what to do about "the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information". Especially, what is "9c5c202"? I guess it might be a link, but I can't click it.

i gusee "9c5c202" is git commit version

[ENGFCC]

Hello, my friend. I meet the same problem as your, did you solve it? Could you please tell me how to make it? I can't know what to do about "the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information". Especially, what is "9c5c202"? I guess it might be a link, but I can't click it.你好，我的朋友。我和你遇到相同的问题，你解决了吗？你能告诉我怎么解决吗？我不知道关于“diff-gaussian-rasterization 的主分支需要修改以补充深度信息”该怎么办。特别是，“9c5c202”是什么意思？我猜这可能是一个链接，但我无法点击它。

i gusee "9c5c202" is git commit versioni gusee "9c5c202" 是 git 提交版本

Thanks for your reply. This problem was solved after I run convert.py again. But, I'm facing a new problem: my "point_cloud.ply" just several MB, far less than author's, and this causes my 3D Editing results very bad, because background is not clear.

[lcicl]

Hello, my friend. I meet the same problem as your, did you solve it? Could you please tell me how to make it? I can't know what to do about "the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information". Especially, what is "9c5c202"? I guess it might be a link, but I can't click it.你好，我的朋友。我和你遇到相同的问题，你解决了吗？你能告诉我怎么解决吗？我不知道关于“diff-gaussian-rasterization 的主分支需要修改以补充深度信息”该怎么办。特别是，“9c5c202”是什么意思？我猜这可能是一个链接，但我无法点击它。

i gusee "9c5c202" is git commit versioni gusee "9c5c202" 是 git 提交版本

Thanks for your reply. This problem was solved after I run convert.py again. But, I'm facing a new problem: my "point_cloud.ply" just several MB, far less than author's, and this causes my 3D Editing results very bad, because background is not clear.

i suggest rebuild the "submodules/diff-gaussian-rasterization" of this branch，and train again

[TesiLin]

I have identified the issue with the rendering result. The current version of the code, when pulled directly, produces a completely black effect.

After my comparison, I found that the main part of the code does not need to be changed. What needs to be changed is the diff -gaussian-rasterization in the submodules . Specifically, in the current branch 9c5c202, the rendered result has depth information, so this version's result is completely black, which is not suitable.

If you switch the submodules/diff Gaussian rasterization to the main branch, the rendered result is correct. Unfortunately, only the depth rendering result is missing.

If you are interested, you can run a comparison to find the final bug in the submodules.

My understanding is that the current version of the code does not match the branch 9c5c202 of diff-gaussian-rasterization marked inside, resulting in almost no significant decrease in loss during rendering.

Either the main body of the code needs to be changed to adapt to the branch 9c5c202 of diff-gaussian-rasterization , or the main body of the code needs not be changed and the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information.

@yzslab @jakubcerveny @emepetres @gdrett

i suggest rebuild the "submodules/diff-gaussian-rasterization" of this branch，and train again

Solved my problem with these two suggestions combined. Many thanks!

[ENGFCC]

I have identified the issue with the rendering result.我已经确定了渲染结果存在的问题。 The current version of the code, when pulled directly, produces a completely black effect.目前版本的代码，直接拉出来是全黑的效果。我已经确定了渲染结果存在的问题。目前版本的代码，直接拉出来是全黑的效果。
After my comparison, I found that the main part of the code does not need to be changed.经过我的比较，我发现代码的主要部分不需要改变。 What needs to be changed is the diff -gaussian-rasterization in the submodules .需要改变的是子模块中的 diff-gaussian-rasterization。 Specifically, in the current branch 9c5c202, the rendered result has depth information, so this version's result is completely black, which is not suitable.具体来说，在当前分支9c5c202中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。渲染结果完全为黑色，问题位于子模块中的 diff-gaussian-rasterization。具体来说，在当前分支 9c5c202 中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。
If you switch the submodules/diff Gaussian rasterization to the main branch, the rendered result is correct.如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。 Unfortunately, only the depth rendering result is missing.不幸的是，只缺少深度渲染结果。如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。不幸的是，只缺少深度渲染结果。
If you are interested, you can run a comparison to find the final bug in the submodules.如果你有兴趣，可以运行比较来找出子模块中的最终错误。如果你有兴趣，可以运行比较来找出子模块中的最终错误。
My understanding is that the current version of the code does not match the branch 9c5c202 of diff-gaussian-rasterization marked inside, resulting in almost no significant decrease in loss during rendering.我的理解是，当前版本的代码与里面标记的diff-gaussian-rasterization分支9c5c202不匹配，导致渲染过程中loss几乎没有明显减少。我的理解是，当前版本的代码与里面标记的 diff-gaussian-rasterization 分支 9c5c202 不匹配，导致渲染过程中 loss 几乎没有明显减少。
Either the main body of the code needs to be changed to adapt to the branch 9c5c202 of diff-gaussian-rasterization , or the main body of the code needs not be changed and the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information.要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。

i suggest rebuild the "submodules/diff-gaussian-rasterization" of this branch，and train again我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练

Solved my problem with these two suggestions combined. Many thanks!解决了我的问题，这两条建议结合在一起。非常感谢！

hello, did you meet some errors when rebuilding the "submodules/diff-gaussian-rasterization" of "3dgs_accel" this branch? I tried it but failed.

[TesiLin]

I have identified the issue with the rendering result.我已经确定了渲染结果存在的问题。 The current version of the code, when pulled directly, produces a completely black effect.目前版本的代码，直接拉出来是全黑的效果。我已经确定了渲染结果存在的问题。目前版本的代码，直接拉出来是全黑的效果。
After my comparison, I found that the main part of the code does not need to be changed.经过我的比较，我发现代码的主要部分不需要改变。 What needs to be changed is the diff -gaussian-rasterization in the submodules .需要改变的是子模块中的 diff-gaussian-rasterization。 Specifically, in the current branch 9c5c202, the rendered result has depth information, so this version's result is completely black, which is not suitable.具体来说，在当前分支9c5c202中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。渲染结果完全为黑色，问题位于子模块中的 diff-gaussian-rasterization。具体来说，在当前分支 9c5c202 中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。
If you switch the submodules/diff Gaussian rasterization to the main branch, the rendered result is correct.如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。 Unfortunately, only the depth rendering result is missing.不幸的是，只缺少深度渲染结果。如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。不幸的是，只缺少深度渲染结果。
If you are interested, you can run a comparison to find the final bug in the submodules.如果你有兴趣，可以运行比较来找出子模块中的最终错误。如果你有兴趣，可以运行比较来找出子模块中的最终错误。
My understanding is that the current version of the code does not match the branch 9c5c202 of diff-gaussian-rasterization marked inside, resulting in almost no significant decrease in loss during rendering.我的理解是，当前版本的代码与里面标记的diff-gaussian-rasterization分支9c5c202不匹配，导致渲染过程中loss几乎没有明显减少。我的理解是，当前版本的代码与里面标记的 diff-gaussian-rasterization 分支 9c5c202 不匹配，导致渲染过程中 loss 几乎没有明显减少。
Either the main body of the code needs to be changed to adapt to the branch 9c5c202 of diff-gaussian-rasterization , or the main body of the code needs not be changed and the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information.要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。

i suggest rebuild the "submodules/diff-gaussian-rasterization" of this branch，and train again我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练

Solved my problem with these two suggestions combined. Many thanks!解决了我的问题，这两条建议结合在一起。非常感谢！

hello, did you meet some errors when rebuilding the "submodules/diff-gaussian-rasterization" of "3dgs_accel" this branch? I tried it but failed.

@ENGFCC No I didn't. What's the errors exactly?

[ENGFCC]

I have identified the issue with the rendering result.我已经确定了渲染结果存在的问题。 The current version of the code, when pulled directly, produces a completely black effect.目前版本的代码，直接拉出来是全黑的效果。我已经确定了渲染结果存在的问题。目前版本的代码，直接拉出来是全黑的效果。
After my comparison, I found that the main part of the code does not need to be changed.经过我的比较，我发现代码的主要部分不需要改变。 What needs to be changed is the diff -gaussian-rasterization in the submodules .需要改变的是子模块中的 diff-gaussian-rasterization。 Specifically, in the current branch 9c5c202, the rendered result has depth information, so this version's result is completely black, which is not suitable.具体来说，在当前分支9c5c202中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。渲染结果完全为黑色，问题位于子模块中的 diff-gaussian-rasterization。具体来说，在当前分支 9c5c202 中，渲染的结果带有深度信息，所以这个版本的结果是全黑的，不太合适。
If you switch the submodules/diff Gaussian rasterization to the main branch, the rendered result is correct.如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。 Unfortunately, only the depth rendering result is missing.不幸的是，只缺少深度渲染结果。如果将子模块/差异高斯光栅化切换到主分支，渲染的结果是正确的。不幸的是，只缺少深度渲染结果。
If you are interested, you can run a comparison to find the final bug in the submodules.如果你有兴趣，可以运行比较来找出子模块中的最终错误。如果你有兴趣，可以运行比较来找出子模块中的最终错误。
My understanding is that the current version of the code does not match the branch 9c5c202 of diff-gaussian-rasterization marked inside, resulting in almost no significant decrease in loss during rendering.我的理解是，当前版本的代码与里面标记的diff-gaussian-rasterization分支9c5c202不匹配，导致渲染过程中loss几乎没有明显减少。我的理解是，当前版本的代码与里面标记的 diff-gaussian-rasterization 分支 9c5c202 不匹配，导致渲染过程中 loss 几乎没有明显减少。
Either the main body of the code needs to be changed to adapt to the branch 9c5c202 of diff-gaussian-rasterization , or the main body of the code needs not be changed and the main branch of diff-gaussian-rasterization needs to be modified to supplement depth information.要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。要么需要改变代码主体以适应 diff-gaussian-rasterization 的分支 9c5c202，要么不需要改变代码主体并且需要修改 diff-gaussian-rasterization 的主分支以补充深度信息。

i suggest rebuild the "submodules/diff-gaussian-rasterization" of this branch，and train again我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练我建议重建此分支的“submodules/diff-gaussian-rasterization”，然后重新训练

Solved my problem with these two suggestions combined. Many thanks!解决了我的问题，这两条建议结合在一起。非常感谢！

hello, did you meet some errors when rebuilding the "submodules/diff-gaussian-rasterization" of "3dgs_accel" this branch? I tried it but failed.

@ENGFCC No I didn't. What's the errors exactly?

Sorry, I forget it. I'll try it again, and then consult you for help, thank you. Are your rendering results good like author's after rebuilding "3dgs_accel" branch?

[TesiLin]

hello, did you meet some errors when rebuilding the "submodules/diff-gaussian-rasterization" of "3dgs_accel" this branch? I tried it but failed.

@ENGFCC No I didn't. What's the errors exactly?

Sorry, I forget it. I'll try it again, and then consult you for help, thank you. Are your rendering results good like author's after rebuilding "3dgs_accel" branch?

@ENGFCC For now, I have tested with tandt/train (mentioned in the paper) and another sample not mentioned in the paper. It seems to be working fine in both rendering and metrics.