Amrita Mazumdar comments

Results 11 comments of


                                            Amrita Mazumdar

lambda_depth=0.0 in arguments/init.py and the role of depth loss during training

Hi, thanks for your question. The depth loss is not used during training, but the MiDaS depth is used to help initialize the COLMAP points (see [this section of the...

lambda_depth=0.0 in arguments/init.py and the role of depth loss during training

For scenes where the initial COLMAP point cloud is sparse, we found that using depth estimation helped with reconstructing that first frame. I don't believe we did experiments comparing different...

Clarification about frame size and pre-trained datasets

Thanks for raising this. You're right that the saved files in the current code release are significantly larger than the sizes reported in the paper. In the paper, we compute...

ValueError: batch_size should be a positive integer value, but got batch_size=0

I have seen this when the dataset_reader does not successfully load training images (you can check this by checking `len(train_image_dataset)` before line 99). Can you check your dataset organization against...

ValueError: batch_size should be a positive integer value, but got batch_size=0

Hi @michnaugh1 , can you paste the full error trace? Where do you get the error exactly? I would expect the code would be able to run on a 4090.

Ubuntu 25.04 Installation error: "ImportError: libtorch_cpu.so: cannot enable executable stack as shared object requires: Invalid argument"

Thanks for raising this issue! We have only tested with Ubuntu 22.04 and GPU Driver Version: 535.129.03, CUDA Version: 12.2 so far. If I understand [this issue with the same...

Ubuntu 25.04 Installation error: "ImportError: libtorch_cpu.so: cannot enable executable stack as shared object requires: Invalid argument"

(sorry, closed by mistake)

Ubuntu 25.04 Installation error: "ImportError: libtorch_cpu.so: cannot enable executable stack as shared object requires: Invalid argument"

I haven't personally tested on a 4090. It does depend on your scene resolution too.

Getting the Gaussians per frame

Hi @szat, you can find the pretrained Gaussians in the [Releases](https://github.com/NVlabs/queen/releases) section. If you follow the instructions to download a single scene (split into tar files for upload size restrictions),...

ungated_xyz_indices.shape[0] == 0

Hi @BruceLz , can you share your run command and config? There may be uncaught logic for saving the compressed pkl if XYZ gating is disabled.