alekceydan / stereo_to_depth

Testing ground for optimal stereo to depth method for 3D stereo.

Implementation takes samples from Inria 3DMovie Dataset (In its current implementation):

The Inria 3DMovie Dataset contains all the stereo pairs and their annotations used in our ICCV 2013 paper [1]. Most of this data was extracted from the "StreetDance 3D" [Giwa and Pasquini, 2010] and "Pina" [Wenders, 2011] stereo movies. Some of the negative stereo pairs were harvested from Flickr and were originally shot using the Fuji W3 camera. The dataset includes stereo pairs, ground truth segmentations, poses and person bounding boxes, and is split into a training and test parts.

References

[2] A. Ayvaci, M. Raptis, S. Soatto
Sparse Occlusion Detection with Optical Flow
In International Journal of Computer Vision, 2011

Copyright

We do not own the copyright of the videos and stereo pairs, and only provide them for non-commercial research purposes.