search

Sharath-girish / Shacira

Official Pytorch implementation of SHACIRA: Scalable HAsh-grid Compression for Implicit Neural Representations
Other
30 stars 3 forks source link

Reproducing Results for the Kodak Dataset from Figure 3. #2

Open nanaimi opened 11 months ago

nanaimi commented 11 months ago

Hi, thanks for releasing the code to your interesting paper!

I would like to reproduce the results in the rate/distortion plot presented in figure 3 of the paper. In the paper you state that varied the per-level entries T, the number of levels L, and the maximum grid resolution. I attempted just scaling the hash table size and achieved good results, but it did not produce the same plot.

Could you give me more insight into how the parameters were scaled to produce the curve?

Thank you!

Sharath-girish commented 11 months ago

Hi, thank you for your interest in our work. We scaled T between 10, 11 and 12 and number of levels L between 16, 20 and 24 with the maximum grid resolution at 512 to generate a similar plot.

darshit-desai commented 9 months ago

I just ran with the default config given in the repo here: https://github.com/Sharath-girish/Shacira/blob/main/app/image/configs/kodak.yaml

It doesn't reproduce the image, what do I need to change in the config?

Sharath-girish commented 9 months ago

Image logging is fixed after commit 37b1a7b There might be small drops in PSNR compared to the paper due to hyperparameter/implementation details. Will update when that is fixed.