I am interested in video processing using (un)supervised deep learning to extract the geometry and semantics of the scenes.
Data for the ICLR2016 submission Spatio-temporal video autoencoder with differentiable memory: dataset_fly_64x64_lines_train.t7 and dataset_fly_64x64_lines_test.t7
Some videos obtained on test real sequences are available here (each frame contains from left to right: ground truth next frame, predicted next frame, and optical flow map).
Library of annotated synthetic indoor scenes SynthCam3D
vp344 at cam.ac.uk
Trumpington Street, Engineering Department, BE4-54, (+44) 01223 765 153