search

mihaibujanca / dynamicfusion

Implementation of Newcombe et al. CVPR 2015 DynamicFusion paper
BSD 3-Clause "New" or "Revised" License
391 stars 105 forks source link

KinFu2 error: invalid configuration argument #70

Open tearsland123 opened 5 years ago

tearsland123 commented 5 years ago

Dear Mihai Bujanca: I am sorry to bother you, I compiled your project on windows and run it via the umbrella database. However, it produced an err like following:​ KinFu2 error: invalid configuration argument cuda/tsdf_volume.cu:829​ I checked your code and found some hints. After the cudaSafeCall ( cudaMemcpyFromSymbol (&size, output_count, sizeof(size)) ) function, size variable will become 0 and it will be returned. I checked issues in github and found someone faced exactly same problem like me. However, there is still no solution related to that up to now.​ If it is possible, could you give me a favor for solving that kind of problem? Thanks for your time.​ ​ Best Regards​ Allen

Algomorph commented 4 years ago

Hi, Mihai,

I'm also facing the same exact issue as @tearsland123. I've debugged through the code a little, and I see that the images of the umbrella dataset seem to be read in just fine by OpenCV. Debugging further, I see that the point cloud that's supposed to be generated from the very first depth image has zero points. I think that is the real issue, since there are definitely valid non-zero depths in the image. This later results in the wrong grid size in the call to extract_normals_kernel.

Perhaps something is clipping away all the points where it shouldn't be? Any ideas?

Thanks in advance for any non-zero input.

mihaibujanca commented 4 years ago

Hi @tearsland123 @Algomorph

My apologies for not responding in time. I'm afraid I've moved on from this project and am quite caught up with other commitments. This project was mainly developed for my own learning and some bits such as Opt were thrown in more for the sake of experimentation than on solid design considerations. I wasn't anticipating that others would take an interest in what I'm developing at the time. As such, there are bottlenecks and bits that are suboptimal, and it also seems like it might not always be easy to get it working.

If you are looking for any open-source non-rigid reconstruction system rather than DynamicFusion specifically, perhaps surfelwarp (https://github.com/weigao95/surfelwarp) could be useful.

Sorry this isn't directly addressing the issue - as I said, I haven't been interacting with the code in quite some time