[Benchmark] Tanks and Temples

Tanks and temples: benchmarking large-scale scene reconstruction
Intel Labs
ACM ToG 2017
[ Download | Paper]

8개의 intermediate scene, 6개의 advanced scene, 7개의 추가 데이터(training data)로 구성됨.

Family	Francis	Horse
Lighthouse	M60	Panther
Playground	Train	Auditorium
Ballroom	Courtroom	Museum
Palace	Temple	Barn
Caterpillar	Church	Courthouse
Ignatius	Meeting room	Truck

Intermediate Scenes

Family

Francis

Horse

Lighthouse

M60

Panther

Playground

Train

• 원본 해상도 : 1920x1080
• 3DGS 해상도 : 980x545

Method	PSNR	SSIM	LPIPS	Training Time	GPU	Storage(MB)	Source
3DGS	21.097	0.802	0.218	-	A6000	-	3DGS
3DGS	22.073	0.821	0.196	14m 05s	RTX3090	259	직접실험

Advanced Scenes

Auditorium

Ballroom

Courtroom

Museum

Palace

Temple

Training Scenes

Barn

Caterpillar

Church

Courthouse

Ignatius

Meeting room

Truck

• 원본 해상도 : 1920x1080
• 3DGS 해상도 : 979x546

Method	PSNR	SSIM	LPIPS	Training Time	GPU	Storage(MB)	Source
3DGS	25.187	0.879	0.148	-	A6000	-	3DGS
3DGS	25.479	0.885	0.142	16m 53s	RTX3090	484	직접실험

Methods

• 3DGS: Train, Truck scene에 대해 수행