Open morsingher opened 3 years ago
作者你好,我在VS2017中运行程序的时候,ACMM和ACMMP都在geometric consistency这部分函数中出现一个Bug,程序直接跳出。并且是在对部分图像完成了Geometric consistancy操作,该Bug尚未找出来
作者你好,我在VS2017中运行程序的时候,ACMM和ACMMP都在geometric consistency这部分函数中出现一个Bug,程序直接跳出。并且是在对部分图像完成了Geometric consistancy操作,该Bug尚未找出来 您好,请问您解决该问题了么
作者你好,我在VS2017中运行程序的时候,ACMM和ACMMP都在geometric consistency这部分函数中出现一个Bug,程序直接跳出。并且是在对部分图像完成了Geometric consistancy操作,该Bug尚未找出来 您好,请问您解决该问题了么
您好!我发现是cuda程序在cuda开辟内存以及指向地址的时候出现---无法读取内存 从而造成程序不进入cuda实现的函数,但还没有找到解决的方法。您知道怎么解决吗?能加我的联系方式吗 QQ 1837379929
Hi @yushengJ , I meet the same issue. It seems that the problem occured in RandomInitialization
.
Run Planar Prior Assisted PatchMatch MVS ...
an illegal memory access was encountered in ~/ACMMP/ACMMP.cu at line 1334
Have you solved this problem? (I searched on QQ but did not find 1837379929, is this number right? @yushengJ )
Hi @yushengJ , I meet the same issue. It seems that the problem occured in
RandomInitialization
.Run Planar Prior Assisted PatchMatch MVS ... an illegal memory access was encountered in ~/ACMMP/ACMMP.cu at line 1334
Have you solved this problem? (I searched on QQ but did not find 1837379929, is this number right? @yushengJ )
Hi, @CanCanZeng! I faced the same issue. Have you solved this problem?
Hi, I think I found a pretty severe bug in both ACMM and ACMMP. The problem is the following line:
https://github.com/GhiXu/ACMMP/blob/f4c78f04b8a78f8565242950190df14a53fba7ab/ACMMP.cu#L1100
When the if statement is not satisfied, that variable remains uninitialized and it is both refined with
PlaneHypothesisRefinement()
and assigned toplane_hypotheses[center]
. This produces random values which then propagates throughout the execution. You can verify it yourself if you just print the number of depth values smaller than 0 and bigger than any reasonable threshold. I would suggest to simply initialize it with the current value:float4 plane_hypotheses_now = plane_hypotheses[center]
The issue doesn't exist in ACMP, where this line is not executed and
plane_hypotheses[center]
is directly manipulated.Now, the problem is that with this fix the 3D output has slightly less points (around 8-9% less), which is weird. Can you look a bit into this issue and clarify what happens? Maybe I'm missing something. Thank you in advance!