ShowMeCode

Hi, I want to use this repo to extract texture, but I got failed when building the main branch. will the `develop branch` solve my problem?

I also have the same problem, the script bash bash_scripts/nerfcapture2dataset.bash configs/iphone/dataset.py always "Waiting for frames...". And I run cyclonedds ps print "Entities discovered: 2", but can not see the NeRFCapture...

I don't understand the effect of the white background, do you mean set white_bkgd=True in the network.py? 

I add mask in gaussian splatting and L1 loss in sugar trainer, which can remove almost background point cloud. but still have some artifacts in background, and holes in mesh.

> @ansj11 Hello, can I ask how to achieve this result? the people in all image is the same pose? 我只是将colmap的点云重新投影到每个视角上，然后根据rembg跑出来的mask判断投影后的uv坐标是否在mask内，不在mask内则丢掉，这样就可以得到人体的点云。只不过在3D GS和sugar优化后，头部和脚部会再引入一些背景点。目前还在解决 

yes，the person is static，i just take a video around the person ------------------ Original ------------------ From: Jett ***@***.***> Date: Wed,Mar 6,2024 8:11 PM To: Anttwo/SuGaR ***@***.***> Cc: ShowMeCode ***@***.***>, Mention ***@***.***>...

> I experience the same issue, have you managed to fix it? no, I didn't use this method. Now I use 2DGS[!https://github.com/hbb1/2d-gaussian-splatting.git]

之前的SD也只做到了64x64的latent特征，再下采样分辨率，扩散生成效果就不好了(虽然不知道为什么)，不过这篇工作看起来又可以了。降低分辨率和加噪感觉还是不一样吧，下采样更像是平滑或者模糊，图像信息还在，只是没了高频信息。另外还有个问题，[1,2,.....,16]只有固定的分辨率，如何向GPT那样运用scaling law呢？

where did you increase the value of -d? I just commented the Assertion code.

> @ansj11 > > ~/Programs/Fotogrametria/openMVS/make/bin/ReconstructMesh **-d 10** --smooth 8 /tmp/testepqp/MVS/scene_dense.mvs thank you for your quick reply, I will have a try!