Iteration start point not working?

[Iliceth]

I have a model trained with gaussian-splatting to 10000 iterations, so it has a folder with the 7000 and 10000 iterations version. I first did it default and it started from the 7000 iterations model as intended.

When I added -i 10000 to python train.py, it clearly uses that input, as my coarse output folder is now called "/sugarcoarse_3Dgs10000_sdfestim02_sdfnorm02" instead of "/sugarcoarse_3Dgs7000_sdfestim02_sdfnorm02".

But it starts from 7000 anyway and goes up to 15000 as with the other test. It the number calculation off and does it actually do 10000 to 18000? Or does it actually not listen to my command?

Start:

Starting regularization...

-------------------
Iteration: 7000
loss: 0.010317  [ 7000/15000] computed in 0.0034049232800801594 minutes.
------Stats-----
---Min, Max, Mean, Std
Points:   -11.007306098937988   16.177127838134766   3.7679994106292725   4.742946624755859
Scaling factors:   1.8295715165095316e-08   0.7787043452262878   0.021300505846738815   0.03534376621246338
Quaternions:   -0.9777704477310181   0.9999950528144836   0.21636688709259033   0.45076122879981995
Sh coordinates dc:   -2.092600107192993   4.579624176025391   -0.21093760430812836   0.833622395992279
Sh coordinates rest:   -0.5350058078765869   0.680113673210144   0.0008770549902692437   0.03914392739534378
Opacities:   0.00291125918738544   0.9999974966049194   0.36435484886169434   0.28180843591690063

End:

-------------------
Iteration: 15000
loss: 0.015563  [15000/15000] computed in 0.4678249955177307 minutes.
------Stats-----
---Min, Max, Mean, Std
Points:   -7.9098005294799805   16.153440475463867   3.7779717445373535   4.672043323516846
Scaling factors:   5.125680324624682e-09   31.3653621673584   0.019177302718162537   0.10984452068805695
Quaternions:   -0.9998881220817566   0.9999532699584961   0.19505058228969574   0.46038711071014404
Sh coordinates dc:   -2.60638427734375   8.151215553283691   -0.14023619890213013   0.8646490573883057
Sh coordinates rest:   -0.7594988346099854   1.1157737970352173   0.0004535722837317735   0.05858450382947922
Opacities:   1.8396954220970853e-15   1.0   0.9946355819702148   0.05587311089038849
Number of gaussians used for sampling in SDF regularization: tensor(23858, device='cuda:0')
Saving model...
Model saved.
Training finished after 15000 iterations with loss=0.01556328497827053.
Saving final model...
Final model saved.