The objective of this project is to evaluate the state-of-the-art on stereo matching algorithms for satellite images. The starting point will be the analysis of a recent CNN-based stereo matching algorithm proposed in [1], or a more recent article judged as more relevant. The context of satellite images is very different from the usual benchmarks used for stereovision, hence the need for a critical analysis. The critical evaluation will focus on the performance on satellite images comparing with techniques currently in use in the domain such as SGM (implemented in the course), MGM [2], variants such as in [3].
Supervision
Gabriele Facciolo, Carlo de Franchis, Enric Meinhardt
Bibliographic References
[1] P. Knobelreiter, C. Reinbacher, A. Shekhovtsov, and T. Pock. End-to-end training of hybrid CNN-CRF models for stereo. CVPR 2017.
[2] G. Facciolo, C. de Franchis, and E. Meinhardt. MGM: A Significantly More Global Matching for Stereovision, BMVC 2015. (python wrapper available in https://github.com/gfacciol/mgm/tree/ctypes)
[3] E. Rupnik, and M. P. Deseilligny. One-Two-Pixel Multi-View Image Matching for digital surface modelling. RFIAP 2018.
