Learning to Estimate Single-View Volumetric Flow Motions without 3D Supervision
Preprint
Abstract
We address the challenging problem of jointly inferring the 3D flow and volumetric densities moving in a fluid from a monocular input video with a deep neural network. Despite the complexity of this task, we show that it is possible to train the corresponding networks without requiring any 3D ground truth for training. In the absence of ground truth data we can train our model with observations from real-world capture setups instead of relying on synthetic reconstructions. We make this unsupervised training approach possible by first generating an initial prototype volume which is then moved and transported over time without the need for volumetric supervision. Our approach relies purely on image-based losses, an adversarial discriminator network, and regularization. Our method can estimate long-term sequences in a stable manner, while achieving closely matching targets for inputs such as rising smoke plumes.Keywords
All Related Versions
This publication has 27 references indexed in Scilit:
- Learning Residual Flow as Dynamic Motion from Stereo Videos2019
- Unsupervised convolutional neural networks for motion estimation2016
- Multi-image 3D reconstruction data evaluationJournal of Cultural Heritage, 2014
- A Quantitative Analysis of Current Practices in Optical Flow Estimation and the Principles Behind ThemInternational Journal of Computer Vision, 2013
- An evaluation of optical flow algorithms for background oriented schlieren imagingExperiments in Fluids, 2008
- An Unconditionally Stable MacCormack MethodJournal of Scientific Computing, 2007
- Density estimation for dynamic volumesComputers & Graphics, 2007
- A Variational Method for Scene Flow Estimation from Stereo Sequences2007
- On 3D scene flow and structure estimation2005
- Whole-field density measurements by 'synthetic schlieren'Experiments in Fluids, 2000