search

YihangChen-ee / HAC

:house: [ECCV 2024] Pytorch implementation of 'HAC: Hash-grid Assisted Context for 3D Gaussian Splatting Compression'
https://yihangchen-ee.github.io/project_hac/
Other
237 stars 16 forks source link

Question about CNC and HAC #16

Open nathanneike opened 1 month ago

nathanneike commented 1 month ago

Hey, Great work first all. I would be curious to know if the method described in CNC has already been implemented in the HAC algorithm (and if not why not).

YihangChen-ee commented 1 month ago

Hi, CNC is partially implemented in HAC.

Specifically, in HAC, the hash grid accounts for less than 5% of the overall model’s storage, so we simply binarize it and apply entropy constraints without incorporating context models to compress the hash grid, as context models on the hash grid would increase training complexity.

In contrast, in CNC, the hash grid comprises the majority of the model’s storage. Therefore, we additionally employ context models to reduce its size as much as possible.

nathanneike commented 1 month ago

Interesting, thank you 👍