Depth from Defocus In the Wild: Addition resultsHuixuan Tang1, Scott Cohen2, Brian Price2, Stephen Schiller2 and Kiriakos N. Kutulakos11University of Toronto 2Adobe Research In CVPR2017 |
Viewing our Global DFD results in 3D |
We export a few of our resulting depth maps to point clouds (.ply). These point clouds can be opened by 3D viewing/editing tools such as Meshlab.
Follow the steps below to view our point clouds in Meshlab: 0. Download and uncompress the PLY(11.6MB) files. 1. Download MeshLab (from sourceforge), install and open it. 2. In Meshlab, click Menu Files->New Empty Project, then cick Menu Files->Import Mesh. 3. Open .ply file in File browser (under ply/). A point cloud should appear in the main window. |
Compared methods |
LDFD (from only 2 images)--our sparse results. GDFD (based on LDFD results)--our dense results. DFF (from 25-41 images)-- Suwajanakorn, S., Hernandez, C., Seitz, S.M.: Depth from focus with your mobile phone. In: Proc. IEEE CVPR. (2015) SPS (based on LDFD results)-- Yao, J., Boben, M., Fidler, S., Urtasun, R.: Real-Time Coarse-to-fine Topologically Preserving Segmentation. In: Proc. IEEE CVPR. (2015) HCF (based on LDFD results)-- Chakrabarti, A., Xiong, Y., Gortler, S.J., Zickler, T.: Low-level vision by consensus in a spatial hierarchy of regions. In: Proc. IEEE CVPR. (2015) |
Results |
Nexus 5:
portrait
bagels
sushi
flower
bell
spike
flower2
christmas
Samsung Galaxy S3:
balls
keyboard
bottles
fruits
plants
telephone
window
(input images and DFF results of Samsung data kinded provided by Suwajanakorn et.al) |
portrait input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() donut input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() flower input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() bell input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() flower6 input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() sushi input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() flower2 input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() christmas input pair (captured by Nexus 5) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() balls input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 25 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() keyboard input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 32 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() bottles input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 41 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() fruits input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 30 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() plants input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 30 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() telephone input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 33 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() window input pair (captured by Samsung Galaxy S3) ![]() LDFD flow ![]() LDFD depth ![]() GDFD flow ![]() GDFD depth ![]() DFF depth (from 27 images) ![]() SPS flow (initialized by LDFD flow) ![]() SPS depth (initialized by LDFD depth) ![]() HCF flow (initialized by LDFD flow) ![]() HCF depth (initialized by LDFD depth) ![]() |