Closed cynthia-you closed 9 months ago
The mesh should be complete. Did you rotate the mesh to viz from a different perspective?
after run 'global_refine' only mesh_cleaned.obj and textured_mesh.obj generated, and i droped the obj in Meshlab.
The mesh should be complete. Did you rotate the mesh to viz from a different perspective?
A, i see, it is becuz every time i run the bundlesdf, the 3d reconsturction results are diffrent:
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that?
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that?
你这里是连续帧出了问题吧,没有匹配上
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that?
你这里是连续帧出了问题吧,没有匹配上
有可能的,但是不管是示例数据集还是自制数据集都频繁发生最后的mesh残缺。
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that? ######
你这里是连续帧出了问题吧,没有匹配上
有可能的,但是不管是示例数据集还是自制数据集都频繁发生最后的mesh残缺。
作者的示例数据集生成的瓶口处是会残缺的,如果别的地方也残缺了,那就是环境可能不对
A, i see, it is becuz every time i run the bundlesdf, the 3d reconsturction results are diffrent:
This is weird. Overall the mesh should be fairly similar. Another weird thing is your texture seems black. Did you try HO3D data as well?
A, i see, it is becuz every time i run the bundlesdf, the 3d reconsturction results are diffrent:
This is weird. Overall the mesh should be fairly similar. Another weird thing is your texture seems black. Did you try HO3D data as well?
Cuze the meash and material_0.png are incompleted, so the mesh looks black on some view, if i rotate the mesh , the missing parts will be more visible.
And in the process of neural field training, the error is very big, the error is up to 1X on rgb_loss and sdf_loss, so the model is incomplete(my assumption).
I havent try HO3D yet.
in many sense, this is abnormal, though I do not have any clues. Did you make any local changes to the code? I'd recommend to start from scratch (cloning the repo, restart docker, etc).
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that?
你这里是连续帧出了问题吧,没有匹配上
有可能的,但是不管是示例数据集还是自制数据集都频繁发生最后的mesh残缺。
######################### another Q ########## When i run dataset 'milk' no Pdb showed, but in my customed dataset , i must manually enter 'c' every epoch, can i do sth to revise that?
你这里是连续帧出了问题吧,没有匹配上
有可能的,但是不管是示例数据集还是自制数据集都频繁发生最后的mesh残缺。
可以私下交流一下吗,我有些问题
Hi 温老师~ After try bundleSDF on GTX1066 ti( loftr always failed), i pulled image on rtx3070, and it run successfully (run_video/global_refine/draw_pose). But the generated 3d mesh(textured_mesh.obj) appears to come from the perspective of a particular frame, not the starting or ending frame. I'm wondering how can i get the completed mesh just like the readme.md showed(the drill demo) , thanks
~