Regarding Training Setting and Dataset Coverage

[shjlin]

Thank you for the very interesting work!
I'm curious about the training setting. In arXiv paper, only DL3DV is mentioned for evaluation. Yet for training, it doesn't seem to have revealed much of the information.
Could you specify on what datasets you used? Appreciate your reply!
Best,
Jonathan

[dragonylee]

Hi Jonathan, thanks for your interest!

We train QuerySplat exclusively on DL3DV-10K (Hugging Face). For evaluation, we use the newer DL3DV-Evaluation benchmark (Hugging Face), which is independent of DL3DV-10K.

The training dataset is also specified in the Implementation Details section of the paper.

Hope this clarifies😄.