1

Light Field Reconstruction Using ConvolutionalNetwork on EPI and Extended Applications

(Supplementary Material)Gaochang Wu, Yebin Liu, Lu Fang, Qionghai Dai, Senior Member, IEEE, and Tianyou Chai, Fellow, IEEE

I. RECONSTRUCTION RESULTS

Bu

dd

ha

Mo

na

Ground truth Wang et al. [39] Jeon et al. [40] Kalantari et al. [8] Ours

Fig. 1. Comparison of light field reconstruction results on the HCI synthetic scenes [45].

Sim

ula

ted

Reference Wang et al. [39] Joen et al. [40] Kalantari et al. [8] Ours

Fig. 2. Comparison of light field reconstruction results on a simulated microscope light field.

2

Rock

Seahors

e

Ground truth Wang et al. [39] Jeon et al. [40] Kalantari et al. [8] Ours

Flo

wer

2F

low

er

2IM

G_1411

IMG

_1328

IMG

_1554

IMG

_1411

IMG

_1328

IMG

_1554

Fig. 3. Comparison of light field reconstruction results on Lytro datasets (1). The datasets are from Kalantari et al.’s dataset [8].

3

II. DEPTH ENHANCEMENT RESULTS

Depth estimation using the input 3×3 views

Depth estimation using the ground truth 9×9 views

Depth estimation using the our reconstructed 9×9 views

Depth estimation using the reconstructed 9×9 views by Kalantari et al. [8]

Buddha Mona

Center views

Horses

Ground truth depth maps

Fig. 4. Depth estimation results using the approach by Wang et al. [39].