Illumination-Computation Acceleration Structure Based on Sparse Voxel Directed Acyclic Graph

Fig. 8. Schematic of acceleration structure simplification based on double depth maps. (a) SVO in space of light rays; (b) SVO after merging adaptively based on node attributes

Download full size

Fig. 9. Optimized SVDAG acceleration structure in space of light rays

Download full size

Fig. 10. Schematic of list corresponding to each layer of SVO sequence in dynamic scene

Download full size

Fig. 11. Schematic of inter-frame multiplexing in SVDAG

Download full size

Fig. 12. Test scenes. (a) Bunny scene; (b) Buddha scene; (c) Dragon scene

Download full size

Fig. 13. Time required for tests of light rays intersection in different acceleration structures

Download full size

Fig. 14. Test scenes. (a) Closed citadel scene; (b) Villa scene; (c) Terrain scene

Download full size

Fig. 15. Relationship between growth factor of storage cost and resolution

Download full size

Fig. 16. Comparison of time costs for different algorithms in Closed citadel scene

Download full size

Fig. 17. Comparison of time costs for different algorithms in Villa scene

Download full size

Fig. 18. Comparison of time costs for different algorithms in Terrain scene

Download full size

Fig. 19. Test scenes. (a) Kitchen scene; (b) Robot scene; (c) Fire scene

Download full size

Table 1. Comparison of total number for nodes and pointers

View table

Table 1. Comparison of total number for nodes and pointers

Scene	Parameter	SVO	ESVO	SVDAG
Bunny	Total number of nodes /10⁴	2307	2543	17.1
Bunny	Total number of pointers /10⁴	529	701	4.8
Buddha	Total number of nodes /10⁴	1577	1837	145.1
Buddha	Total number of pointers /10⁴	438	487	43.4
Dragon	Total number of nodes /10⁴	890	221	14.2
Dragon	Total number of pointers /10⁴	214	53	3.8

Table 2. Comparison of storage costs

View table

Table 2. Comparison of storage costs

Scene	Resolution /(pixel×pixel×pixel)	Storage cost /MB
Scene	Resolution /(pixel×pixel×pixel)			SVO	ESVO	SVDAG
Bunny	512×512×512	5.82	31.93	1.06
Bunny	1000×1000×1000	22.42	123.50	3.32
Buddha	512×512×512	3.49	36.57	7.44
Buddha	1000×1000×1000	13.94	126.60	26.30
Dragon	512×512×512	0.25	0.47	0.08
Dragon	1000×1000×1000	0.82	1.49	0.31

Table 3. Comparison of build time of acceleration structures

View table

Table 3. Comparison of build time of acceleration structures

Resolution /(pixel×pixel×pixel)	Bunny		Buddha		Dragon
Resolution /(pixel×pixel×pixel)	$\begin{matrix} {t_{A}}_{} \end{matrix}$ /ms	$\begin{matrix} {t_{B}}_{} \end{matrix}$ /ms	$\begin{matrix} {t_{A}}_{} \end{matrix}$ /ms	$\begin{matrix} {t_{B}}_{} \end{matrix}$ /ms	$\begin{matrix} {t_{A}}_{} \end{matrix}$ /ms	$\begin{matrix} {t_{B}}_{} \end{matrix}$ /ms
128×128×128	6.14	2.29	8.75	3.06	11.36	4.31
256×256×256	24.06	5.07	33.55	8.07	40.25	11.30
512×512×512	250	30	340	40	400	50
1000×1000×1000	1020	130	1420	190	1600	220
2000×2000×2000	4250	490	5800	780	6530	890

Table 4. Comparison of storage costs for different algorithms in different scenes

View table

Table 4. Comparison of storage costs for different algorithms in different scenes

Resolution /(pixel×pixel×pixel)	Closed citadel			Villa			Terrain
Resolution /(pixel×pixel×pixel)	S₁ /MB	S₂ /MB	S₃ /MB	S₁ /MB	S₂ /MB	S₃ /MB	S₁ /MB	S₂ /MB	S₃ /MB
1000×1000×1000	2.30	0.493	0.231	2.30	0.462	0.146	2.30	0.592	0.538
2000×2000×2000	9.27	1.62	0.532	9.27	2.47	0.336	9.27	2.37	2.03
4000×4000×4000	37.20	5.10	1.23	37.20	7.78	0.772	37.20	7.50	6.64
16000×16000×16000	605.72	47.47	6.46	605.72	85.29	4.09	605.72	91.20	81.59
32000×32000×32000	2422	127.14	13.87	2422	249.04	9.40	2422	183.23	163.92

Table 5. Average number of nodes per frame and total number of nodes

View table

Table 5. Average number of nodes per frame and total number of nodes

Scene	SVO		Difference tree		SVDAG		Time-correlation SVDAG
Scene	$\begin{matrix} {N_{p}}_{} \end{matrix}$ /10³	$\begin{matrix} {N_{n}}_{} \end{matrix}$ /10⁶	$\begin{matrix} {N_{p}}_{} \end{matrix}$ /10³	$\begin{matrix} {N_{n}}_{} \end{matrix}$ /10⁶	$\begin{matrix} {N_{p}}_{} \end{matrix}$ /10³	$\begin{matrix} {N_{n}}_{} \end{matrix}$ /10⁶	$\begin{matrix} {N_{p}}_{} \end{matrix}$ /10³	$\begin{matrix} {N_{n}}_{} \end{matrix}$ /10⁶
Kitchen	873	685	15.1	1.71	94.8	6.87	5.82	0.36
Robot	321	68.7	224	56.6	50.7	10.0	29.3	7.28
Fire	593	192	433	158	62.1	20.8	44.0	13.66

Table 6. Rendering performance of test scenes
View table
Table 6. Rendering performance of test scenes
Scene $\begin{matrix} {V_{1}}_{} \end{matrix}$ /(frame·s^-1) $\begin{matrix} {V_{2}}_{} \end{matrix}$ /(frame·s^-1) $\begin{matrix} {V_{3}}_{} \end{matrix}$ /(frame·s^-1)
Kitchen 21.3 28.1 62.9
Robot 12.0 16.1 53.4
Fire 6.20 9.90 27.1

Tools

Get Citation

Copy Citation Text

Yuwei Yuan, Jicheng Quan, Chen Wu, Yu Liu, Hongwei Wang. Illumination-Computation Acceleration Structure Based on Sparse Voxel Directed Acyclic Graph[J]. Acta Optica Sinica, 2017, 37(8): 0820001

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites