search

zju3dv / PGSR

code for "PGSR: Planar-based Gaussian Splatting for Efficient and High-Fidelity Surface Reconstruction"
https://zju3dv.github.io/pgsr/
Other
403 stars 22 forks source link

Out of memory when evaluating DTU mesh #15

Closed ingra14m closed 1 month ago

ingra14m commented 1 month ago

Hi, thanks for the great work!

I found that the reconstructed mesh (around 500MB) in DTU dataset has too many faces, which will lead to oom errors upon executing radius_neighbors.

My computer has 32GB RAM and RTX 3090 with 24GB VRAM. I have no idea if this problem stems from the memory or something else I have ignored. Tks so much for your help!

downsample pcd: 33%|█████████████████████████ | 3/9 [05:31<10:41, 106.98s/it] [1] 3137183 killed python scripts/dtu_eval.py --data --scan 69 --mode mesh --dataset_dir

danpeng2 commented 1 month ago

Hello, thank you for your interest in our work.

The reconstructed Mesh size from the DTU scan69 sequence is approximately 60M. May I ask if your DTU dataset includes mask? Has this line of code been executed? Could you show the reconstructed Mesh image?

ingra14m commented 1 month ago

Hi, thanks for the prompt reply.

I have identified the problem, which stemmed from an incorrect method of generating the mesh used for evaluation.

Initially, I used python render.py -m path, which generated the mesh without a mask.

When I used python scripts/render_dtu.py -m path, the evaluation code ran successfully.