Evaluation

We evaluate our implementation of Gaussian Splatting (Splatfacto) on the Mip-NeRF 360 dataset, benchmarking it against the original Inria method. We evaluate all methods with the same resolution (2x downscale) and COLMAP params/undistortion, and we report results at 7,000 and 30,000 steps. All evaluations were executed on an NVIDIA RTX 4090 GPU.

Time

Bicycle

Bonsai

Counter

Flowers

Garden

Kitchen

Stump

Treehill

Avg

inria-7k

3:34

3:27

3:05

3:02

3:51

4:00

3:25

2:53

3:24

splfacto-7k

2:36

2:18

2:06

2:14

2:49

2:17

2:24

2:23

2:23

splfacto-big-7k

3:41

3:11

2:59

2:51

3:51

3:15

2:49

2:47

3:10

inria-30k

25:07

14:37

15:24

17:32

24:03

19:02

20:25

18:02

19:17

splfacto-30k

18:03

10:13

9:07

13:25

14:58

10:02

15:15

16:56

13:30

splfacto-big-30k

31:05

13:32

13:27

22:44

26:03

16:16

21:58

25:26

21:19

PSNR

Bicycle

Bonsai

Counter

Flowers

Garden

Kitchen

Stump

Treehill

Avg

inria-7k

24.11

29.49

27.16

20.54

26.53

29.02

26.74

22.50

25.76

splatfacto-7k

22.99

29.45

26.92

20.33

25.76

28.48

24.59

21.91

25.05

splatfacto-big-7k

23.66

29.69

27.01

20.73

26.58

28.82

25.69

22.11

25.54

inria-30k

25.61

31.89

28.96

21.56

27.60

31.30

25.89

22.07

26.86

splatfacto-30k

24.99

32.14

28.72

21.54

27.31

31.18

25.64

22.28

26.73

splatfacto-big-30k

25.7

32.23

28.95

21.96

27.83

31.6

26.7

22.38

27.17

LPIPS

Bicycle

Bonsai

Counter

Flowers

Garden

Kitchen

Stump

Treehill

Avg

inria-7k

0.31

0.24

0.25

0.42

0.16

0.16

0.28

0.42

0.28

splfacto-7k

0.31

0.16

0.21

0.44

0.15

0.14

0.28

0.45

0.27

splfacto-big-7k

0.28

0.16

0.20

0.42

0.12

0.13

0.23

0.43

0.24

inria-30k

0.21

0.21

0.20

0.34

0.11

0.13

0.22

0.32

0.22

splfacto-30k

0.18

0.13

0.17

0.34

0.09

0.10

0.18

0.32

0.19

splfacto-big-30k

0.15

0.13

0.15

0.31

0.07

0.09

0.15

0.28

0.17

SSIM

Bicycle

Bonsai

Counter

Flowers

Garden

Kitchen

Stump

Treehill

Avg

inria-7k

0.69

0.92

0.88

0.53

0.83

0.90

0.73

0.59

0.76

splfacto-7k

0.65

0.92

0.88

0.53

0.85

0.90

0.68

0.58

0.74

splfacto-big-7k

0.69

0.92

0.88

0.55

0.84

0.90

0.74

0.61

0.77

inria-30k

0.78

0.94

0.91

0.61

0.87

0.92

0.77

0.63

0.80

splfacto-30k

0.75

0.94

0.90

0.60

0.85

0.92

0.73

0.63

0.79

splfacto-big-30k

0.78

0.94

0.91

0.63

0.88

0.93

0.77

0.64

0.81