Sciweavers

30
Voted
CVPR
2006
IEEE

Learning Joint Top-Down and Bottom-up Processes for 3D Visual Inference

14 years 3 months ago
Learning Joint Top-Down and Bottom-up Processes for 3D Visual Inference
We present an algorithm for jointly learning a consistent bidirectional generative-recognition model that combines top-down and bottom-up processing for monocular 3d human motion reconstruction. Learning progresses in alternative stages of self-training that optimize the probability of the image evidence: the recognition model is tunned using samples from the generative model and the generative model is optimized to produce inferences close to the ones predicted by the current recognition model. At equilibrium, the two models are consistent. During on-line inference, we scan the image at multiple locations and predict 3d human poses using the recognition model. But this implicitly includes one-shot generative consistency feedback. The framework provides a uniform treatment of human detection, 3d initialization and 3d recovery from transient failure. Our experimental results show that this procedure is promising for the automatic reconstruction of human motion in more natural scene set...
Cristian Sminchisescu, Atul Kanaujia, Dimitris N.
Added 10 Jun 2010
Updated 10 Jun 2010
Type Conference
Year 2006
Where CVPR
Authors Cristian Sminchisescu, Atul Kanaujia, Dimitris N. Metaxas
Comments (0)