Significant difference in Gaussian count

[Walker367]

Thank you for your excellent work.
When trying to reproduce the Full-body Avatar part in SplattingAvatar, I found that the number of generated Gaussian count was much less than that in the checkpoints provided in the project.

1. Following the project documentation, I correctly installed all the project dependencies.
2. I downloaded the PeopleSnapshot’s parameter files provided in the readme.
3. I ran the project training script python train_splatting_avatar.py --config "configs/splatting_avatar.yaml;configs/instant_avatar.yaml" --dat_dir dataset/peoplesnapshot/female-3-casual, and the number of Gaussian in the training results was much less than the normal number shown in the checkpoints.
   The content of the stats.json file output by my project run is as follows:
   {"psnr": 30.43978620042988, "ssim": 0.9768667022387186, "lpips": 0.04340018843318902, "n_gauss": 88614}