Computational Mirrors: Blind Inverse Light Transport by Deep Matrix Factorization
This work addresses the challenge of blind inverse light transport for hidden scene reconstruction, presenting a novel computational method.
The authors tackled the problem of recovering hidden scene motion from indirect illumination changes by factorizing observed video into unknown hidden scene video and light transport matrices, using randomly initialized CNNs to achieve decompositions that reflect true motion.
We recover a video of the motion taking place in a hidden scene by observing changes in indirect illumination in a nearby uncalibrated visible region. We solve this problem by factoring the observed video into a matrix product between the unknown hidden scene video and an unknown light transport matrix. This task is extremely ill-posed, as any non-negative factorization will satisfy the data. Inspired by recent work on the Deep Image Prior, we parameterize the factor matrices using randomly initialized convolutional neural networks trained in a one-off manner, and show that this results in decompositions that reflect the true motion in the hidden scene.