flamehaze1115
commented
1 year ago Hello. The background has surfaces is a common problem for neural rendering-based methods, including neus or volsdf. This is because only images are utilized as supervision. For texture-less background, any surface predictions will be acceptable, since the regions will have very small errors in the rendering loss. This is why we propose a consistence-aware FT to enhance the predicted model by generic model and remove the free surfaces of background parts.
Modifying this parameter will only influence the color of the synthesized images.
If you want to make the generic model to predict clean surfaces of the target object, you have to preprocess the input images to mask out its background, and enforce the background part to be empty in the training.
In the figures of paper, we just use a smaller bounding box to cut off the predicted mesh of the generic model to only keep the foreground part. If you train the generic model on a larger dataset not just one scene, the surfaces of the foreground and background will be separate like the first figure you plotted in this issue.
MNILjj
zjhthu
Dear author,
Thanks for your excellent work! I want to try to get model without the background,I downloaded your training set and remove the background. No matter whether "use_white_bkgd" is set to "True" or "False", the obtained model still has background. So I need to modify some parameters in the "general_lod0.conf" file?
The context I'm talking about is the part outside the predictive model, as shown in the image below:
The parameter I modified is this place:
This is the training image and the predicted model after I removed the background. I have tried both white and black backgrounds, and the effect is not very good.
In order to quickly verify my ideas, currently I only train scan7 on one data. Excuse me, what do I need to do if I need to make the network only predict the model？thanks.