search

BaowenZ / RaDe-GS

RaDe-GS: Rasterizing Depth in Gaussian Splatting
Other
460 stars 25 forks source link

numpy.core._exceptions._ArrayMemoryError #22

Open hefan14 opened 3 months ago

hefan14 commented 3 months ago

2 Hello. When I ran the evaluation code, the following errors occurred: numpy.core._exceptions._ArrayMemoryError:Unable to allocate 19.1MiB for an array with shape (1537052,) and data type [('vertex_indices', [('f0', 'u1'), ('f1', '<u4', (3, ))])]. I don't understand why it occupies up to 64GB of memory. I would like to know if you have encountered such a problem before.

BaowenZ commented 3 months ago

The function mesh.split costs much memory. Please comment it out and directly output mesh in cull_mesh. And I'll replace it with another function.

hefan14 commented 3 months ago

Thank you, it has been a great help to me！

hefan14 commented 3 months ago

May I ask what the specific results are represented by the output of 'vis/vis_201_d2s. ply' and 'vis/vis_201_s2d_ply' in 'eval. py'? How should I obtain valid information from them?

hefan14 commented 3 months ago

1 2 Hi! The grid results I obtained are not as perfect as those provided on the 'Project page'. Can you provide some possible reasons or suggestions?

BaowenZ commented 3 months ago

Hi! 'vis/vis_201_d2s. ply' and 'vis/vis_201_s2d_ply' represent the closest distances between output and ground truth. Red parts indicate large error. And please check that you have reinstall the package with the right version. And you could print the rendered images during training to check training process.

hefan14 commented 3 months ago

I retrieved the latest version of the code and conducted the training process, but unfortunately, the grid results I extracted were not ideal and differed significantly from the results provided on the project page.