When using evalglare to calculate the DGP, it costs about 2 minutes for one image. If annal images were need to be calculated, it always takes 2 hours. Even if we used multithread to accelerate the run, it also need to take ten minutes. I read the source code of the evalglare, and it seems that a annual glare calculation method has been updated, but I don’t know how to use it.
Hi Yongqing,
if you are ready to rethink a bit how the simulations are done, you can
do annual glare evaluations also for only one sensor point and direction
with /Raytraverse/ (although that software was primarily developed to
accound for entire zones with many view directions). /Raytraverse/
replaces the image-based simulation by an adaptive, directionally sparse
calculation. This also allows faster evaluation in terms of glare
metrics than full image-based solutions, but avoids the shortcomings of
e.g. illuminance-based proxies:
Wasilewski, Grobe, Wienold, & Andersen (2022). Efficient simulation for
visual comfort evaluations. /Energy and Buildings/ 267.
https://doi.org/10.1016/j.enbuild.2022.112141
The documentation of the latest release is at
https://raytraverse.readthedocs.io
and the use of the software was introduced e.g. in a tutorial last year
at the Radiance Workshop in Toronto:
https://www.radiance-online.org/community/workshops/2022-toronto-canada
Best, Lars.
As Lars and Nathaniel said there were efficient solutions for annual glare simulations developed in the last years - especially raytraverse combines efficiency and accuracy.
Stephen Wasilewski compared many methods regarding accuracy and speed here: https://doi.org/10.1051/e3sconf/202236201002 (" A Critical Comparison of Annual Glare Simulation Methods").
What you saw in the source code of evalglare belongs probably to the (not yet published) successor of the eDGPs method (called AGC “Adaptive Glare Coefficient” ), that was also tested in the same paper. This piece of code can’t help to speed up when you have a fully rendered image - it is meant to evaluate only parts of an image, which is one of the features of the AGC. The AGC-method could run a full annual simulation for one point with pre-calculated illuminances e.g. in less than 2min using a simplified fabric model (that was wrongly called “roos-wienold” in a paper from Taoning, ignoring the co-authorship of T. Kuhn and J. Christoffersen. Will be probably called differently later). The AGC as tool is at the moment only in experimental stage and not yet bug-free.
One way to speed up evalglare for your situation is to reduce the image size e.g. to 700x700 (using pfilt -1 -x 700 -y 700 inp.hdr |evalglare… ) This low resolution is in my point of view ok since you are downsampling and therefore you should not miss a peak - rendering from the beginning with a low resolution has a high risk to miss a peak that can cause glare.
Jan