Hello, I’ve noticed that the quality of the outputs I’m getting from the latest version of the high-quality generation model is significantly lower compared to the older version. Additionally, there are artifacts in the generated results. I’m not sure if there’s something wrong with my settings. Can you help me with this?
Hey, sorry to hear about the quality degradation. Which previous version were you using? There were no changes in model reconstruction between version 1.3.2 and 1.4. Screenshots of your settings from both versions will also be helpful.
Possibly, in the Reconstruction settings, check which Depth map algorithm version are you using. Version 2 is much faster but might loose finer details in some datasets.
Hello, I am using a version below 1.3. After testing, I found the quality of version 2 to be lower than version 1. It will be slower, but for me, I need quality. Do you know what the problem is with the noise caused by the baking texture in Figure 1?
Hello, what parameters should I adjust if I need to do high-quality calculations？
OK, in that case. Switch to the Version 1 depth map algorithm in the Reconstruction Settings.
Regarding the issue in the texture reprojection, is this something happening only in 1.4 or can you reproduce the same issue with the same data in the earlier version?
In the meantime, try the following:
Sometimes, the projection distance won’t project far enough. Try changing it from Automatic to Custom and increase the reprojection distance.
Changing the Supersampling from automatic to 64 samples per texel can increase the quality, but processing will take longer.
The Texture Sampling method in the Color reprojection settings set to Trilinear can produce smoother results than Nearest but can take longer.
Let me know how did it go for you.
Jakub
The texture reprojection issue did not occur in the version 1.0.3 that I used, and I did not utilize the texture projection feature. The texture reprojection problem only appeared after updating to the new version. Here is my automation code. Could you please review this process and let me know if there are any issues? Based on the current computational results, the quality of version 1 is not as good as that of version 1.0.3. Is there a better configuration or setting to achieve the optimal model quality? This would enable us to provide artists with free and high-quality assets. I look forward to your response.
Wow, what an old version of RealityCapture. As far as I know, 1.0.3 hard expired long time ago and should not be working for years now. I wonder how you made it work 
Anyway, add the following line before -mvsHigh
Add
-set "MvsDepthMapsLibVersion=0"
This will use the old depth map library.
You will have to experiment with the other settings to achieve the best results. The default values are made to work in the majority of cases.
Well, I’ll run some more tests to see if there is any improvement. Thank you for your advice.