Cannot get similar results as provided checkpoints

[Wang-Wenqing]

Thanks for your great work. I encounter trouble om reproducing results, I followed the preprocess_PeopleSnapshot.py get the peoplesnapshot data, and the training command, and cannot find the problem, so could you give some advice about what could be the cause?

I trained the male-4-casual eval_30000 results like this,
{"psnr": 30.53088405397203, "ssim": 0.9789141797357135, "lpips": 0.04090929890258445, "n_gauss": 104333}
And you provided checkpoint have results like this:
{"psnr": 30.9997394879659, "ssim": 0.9815233714050717, "lpips": 0.029089721178428993}

And also the animation has more obvious visialization difference, for example,
this is mine trained model results,

and this is you provided checkpoint results,

[foooolish]

Hi, the model I trained has no material, how do I add a material?

[superhero177]

@Wang-Wenqing
Hello, I encountered a problem similar to yours, but my metrics are much worse. I haven’t made any changes to the code, so I was wondering if you have made any adjustments? If it’s convenient, could you please offer some guidance? Thank you very much!
{"psnr": 25.030534532335068, "ssim": 0.9597002648644977, "lpips": 0.06380893590135707, "n_gauss": 282032}