Which pretrained model did you use on the DND benchmark?

[hypercost]

It is reported in your paper that

To this end, we evaluate on the recent Darmstadt Noise Dataset [34], consisting of 50 noisy images shot with four different cameras at varying ISO levels. Realistic noise can be well explained by a Poisson-Gaussian distribution which, in turn, can be well approximated by a Gaussian distribution where the variance depends on the image intensity via a linear noise level function [12].

So the results_poissongaussian_denoising/pretrained is the model you used on DND dataset?

[hypercost]

Hi, I can't run evaluation on DND dataset on one 1080Ti even with TC. By the way, when I install Tc with conda, the pytorch will downgrade to 0.3.1.

[tobiasploetz]

Hi,

yes, results_poissongaussian_denoising/pretrained is the model that reproduces the DND benchmark results.

I can't run evaluation on DND dataset on one 1080Ti even with TC. By the way, when I install Tc with conda, the pytorch will downgrade to 0.3.1.

This is a bit unfortunate, indeed. I think there are two ways of handling this situation.

1. Build TC by yourself (see here).

2. Have another python environment with TC and pytorch 0.3.1. The code for evaluation should be mostly compatible with this pytorch version.

Ideally I want to get rid of the TC dependency and instead have a cuda kernel that implements the functions indexed_matmul_1 and indexed_matmul_2 directly. However, I haven't had time for this so far :(

Best, Tobias