HashDAG
HashDAG copied to clipboard
Other Voxel Scenes?
Hello,
I am highly interested in the HashDAG approach and was able to run the Windows demo that used Epic Citadel and I was wondering what other types of scenes could be used as I am still trying to get a feel for things?
In particular, I am interested in what type of files could be loaded and what limitations there might be on this.
Any guidance would be greatly appreciated.
Thanks
I just found some on your Google Drive:
https://drive.google.com/open?id=1P3T_wlC3vOcxQSKh9QD0WtB3yNlbPLV2
So then, my next question would be as to how those are created although there is some discussion on using DAG_Compress:
https://github.com/Phyronnaz/DAG_Compression
But that seems to be still under development.
Can other software create the voxel scenes that are then compressed with the DAG_Compressor?
I have the data files for Sponza Sanmiguel and others now but there seems to be no information on how to run them and I have only run the Epic Citadel with success.
Please let me know the proper way to run the other data sets that are in 64K, and 32K resolution as well.
Thanks
Hello,
Is there any update on this?
I am trying to load up other scenes instead of just the Epic Citadel but things are not working out so far.
What settings to I need to change? Thanks
- The settings are in typedefs.h. See e.g., SCENE_NAME and SCENE_DEPTH and similar.
- It's unlikely that there is other software that outputs compatible files (there might be some other software that applies the SVDAG compression, but they are likely using their own file format(s)). You can try with the original fork of the DAG_Compression app, which seems to have been updated more recently. It's linked from the repo that you mention above
FWIW- the storage format of the data isn't that complicated and the compression scheme is described across a number of papers. See e.g., https://diglib.eg.org/handle/10.2312/egt20181028 and papers referred therein.
Hello,
I actually figured it out and had been setting the SCENE and SCENE_DEPTH in the "typedefs.h"
SCENE_DEPTH 15 REPLAY_DEPTH 15 SCENE "sponza"
Turns out that you get a HUGE slowdown if you build with Visual Studio 2019 and CUDA 10.2 in Debug mode with rendering in the 150 ms range while if you build and run in "Release" mode then you get reasonable performance at about 6 ms. (Trace Primary)
I am wondering if you plan on updating to the latest CUDA 12.1 or perhaps Vulkan although I am not sure how much gain/loss that would give in performance.
Also, I did send you an email with to possible discuss some ideas on "Real-Time Polygonal-Light Shading with Linearly Transformed Cosines" (https://eheitzresearch.wordpress.com/415-2/) that might be better than the current ray-tracing approach that you are using in HashDAG now but was not sure you had received it. I am really interested in adding effective rendering for an eventual photorealistic system, one day should the technology advance enough.
While researching HashDAG more I came across "Lossy Geometry Compression for High Resolution Voxel Scenes" https://graphics.tudelft.nl/Publications-new/2020/VSE20/ which might be of interest to you as well.
Lastly, as this current HashDAG work is dated 3 years ago, I was wondering if there have been any advancements around this framework or updated publications that I could get to review.
Thanks again
Hello,
Just wanted to see if you had received this inquiry?
Also, did you use the "DAG-Example" (https://github.com/gegoggigog/DAG-example/tree/compression) to create the other models found at:
https://drive.google.com/open?id=1P3T_wlC3vOcxQSKh9QD0WtB3yNlbPLV2
Thanks
Any updates on how you created the scenes?
Is there going to be no support for these questions?
I just need to know how you created the 64K^3 and 128K^3 scene files that you have on your Google Drive so that I can create others.
What tools and what steps did you use to create a typical scene and color file like
epiccitadel128k.basic_dag.compressed_colors.variable.bin and epiccitadel128k.basic_dag.dag.bin
Please clarify the steps so that I can reproduce here. Thanks