Clarification on Resolution Used in HyperNeRF for vrig Scenes

[sandokim]

Hello, great work on the 4DGS project!

I’ve been exploring the HyperNeRF dataset and wanted to clarify a few points regarding the vrig scenes. When I downloaded the dataset from HyperNeRF, I found the following:

hypernerf_vrig

• vrig-chicken, vrig-3dprinter, broom:

  • 1x Folder: Images at 1072x1920 resolution
  • 2x Folder: Images at 536x960 resolution.

• peel-banana:

  • 1x Folder: Not available.
  • 2x Folder: Images at 536x960 resolution.

The dataset seems to include 4 scenes for the vrig dataset, and for these scenes, the original resolution for most images is 1072x1920, with 536x960 as the half-resolution.

I wanted to ask: Did 4DGS use only the vrig scenes for their experiments?

Additionally, in 4DGS Table 2, the resolution is listed as 960x540, but I was wondering if this might actually be 960x536 instead for all the methods in Table 2 (Nerfies, HyperNeRF, TiNeuVox-B, 3D-GS, FFDNeRF, V4D, 4DGS(Ours)),

considering the resolution of hypernerf dataset downloaded from Hypernerf dataset as follows:

Lastly, was there any discussion or exchange of information regarding the hypernerf dataset resolution between the 4DGS authors and the HyperNeRF authors? It would be great to understand if there were any changes or clarifications on this topic.

Thanks in advance for your help, and keep up the amazing work!

[sandokim]

I’ve been reviewing the hyper_loader.py file in the 4DGS codebase and noticed that it loads imgs, depths, and masks from the 2x folder without any resolution adjustment.

https://github.com/hustvl/4DGaussians/blob/master/scene/hyper_loader.py#L96

However, the images in that folder from HyperNeRF have a resolution of 536x960.

I apologize if this is a naive question, but I wanted to ask whether this height discrepancy is expected and if there's any implicit handling (e.g., cropping or padding) elsewhere in the pipeline. Clarifying this would be very helpful for understanding how to prepare the hypernerf vrig data correctly.

Thank you!