The KITTI Vision Benchmark Suite

This paper proposes a novel permutation
formulation to the stereo matching problem. Our
proposed approach introduces a permutation volume
which provides a natural representation of stereo
constraints and disentangles stereo matching from
monocular disparity estimation. It also has the
benefit of simultaneously computing disparity and
a confidence measure which provides explainability
and a simple confidence heuristic for occlusions.
In the context of self-supervised learning, the
stereo performance is validated for standard
testing datasets and the confidence maps are
validated through stereo-visibility. Results show
that the permutation volume increases stereo
performance and features good generalization
behaviour. We believe that measuring confidence is
a key part of explainability which is instrumental
to adoption of deep methods in critical stereo
applications such as autonomous navigation.

Our model is trained on the datasets at half
resolution on random image crops of size of 192 ×
32 pixels with a batch size of 2. No other data
augmentation is applied apart from random crops.
The conv blocks are as defined in [18] and their
Fig. 5 (right) with f=32. The convolution layers
apply fixed padding, have batch normalization and
have a Relu activation function. The
implementation is made with Mathematica[21] 12.3.
The λ is set to 10 and the symmetric normalization
has t=8 iterations. Networks are trained until
convergence with the Adam Optimizer[24] and a
learning rate of 1×10−3. The constant α is set to
0.85 as is customary [7] and τ is set to 0.1. The
models are trained on an RTX3090.

@inproceedings{brousseau2022permutation,
title={A Permutation Model for the Self-
Supervised Stereo Matching Problem},
author={Brousseau, Pierre-Andr{\'e} and Roy,
S{\'e}bastien},
booktitle={2022 19th Conference on Robots and
Vision (CRV)},
pages={122--131},
year={2022},
organization={IEEE}
}