Hi. Thanks for your Excellent Work.
When attempting to reproduce the results from the Our-3D-GS table, I found that I was unable to replicate the results as shown on NeRF-Synthetic and TandT datasets
Specifically, I used the script below, adjusting only the parameters related to the dataset path.
python train.py --config config/explicit/3D/lod_sh2_mean.yaml
get tandt/train datasets result as shown:
but I find that i didn't meet such problem on mipnerf datasets (at least not such problem on mipnerf/bicycle datasets)：
I also find that I can't reproduce good result on NeRF-Synthetic datasets. with using both the scirpt shown below:
python train.py --config config/explicit/3D/lod_sh2_mean.yaml
python train.py --config config/implicit/3D/lod_sh2_mean.yaml
I believe it is partly from the alpha_mask when handling the NeRF-Synthetic datasets as the issuse: https://github.com/graphdeco-inria/gaussian-splatting/issues/1038
However, after commenting out the code related to alpha_mask in the training script, I still couldn't achieve good results on the NeRF-Synthetic datasets.
Could you guide me on how to fix this issue?
By the way, when attempting to reproduce the results from another one of your repositories, Octree-GS, I didn’t encounter any issues on both NeRF-Synthetic datasets and TandT datasets.
Hi. Thanks for your Excellent Work.
When attempting to reproduce the results from the Our-3D-GS table, I found that I was unable to replicate the results as shown on NeRF-Synthetic and TandT datasets
Specifically, I used the script below, adjusting only the parameters related to the dataset path.
python train.py --config config/explicit/3D/lod_sh2_mean.yamlget tandt/train datasets result as shown:
but I find that i didn't meet such problem on mipnerf datasets (at least not such problem on mipnerf/bicycle datasets)：
I also find that I can't reproduce good result on NeRF-Synthetic datasets. with using both the scirpt shown below:
python train.py --config config/explicit/3D/lod_sh2_mean.yamlpython train.py --config config/implicit/3D/lod_sh2_mean.yamlI believe it is partly from the alpha_mask when handling the NeRF-Synthetic datasets as the issuse: https://github.com/graphdeco-inria/gaussian-splatting/issues/1038
However, after commenting out the code related to alpha_mask in the training script, I still couldn't achieve good results on the NeRF-Synthetic datasets.
Could you guide me on how to fix this issue?
By the way, when attempting to reproduce the results from another one of your repositories, Octree-GS, I didn’t encounter any issues on both NeRF-Synthetic datasets and TandT datasets.