Practical Deep Stereo (PDS): Toward applications-friendly deep stereo matching

Authors: Stepan Tulyakov, Anton Ivanov, François Fleuret

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare PDS to state-of-the-art methods published over the recent months, and demonstrate its superior performance on Flying Things3D and KITTI sets.
Researcher Affiliation Academia Stepan Tulyakov Space Engineering Center at École Polytechnique Fédérale de Lausanne stepan.tulyakov@epfl.ch Anton Ivanov Space Engineering Center at École Polytechnique Fédérale de Lausanne anton.ivanov@epfl.ch Francois Fleuret École Polytechnique Fédérale de Lausanne and Idiap Research Institute francois.fleuret@idiap.ch
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No We guarantee reproducibility of all experiments in this section by using only available data-sets, and making our code available online under open-source license after publication.
Open Datasets Yes We used three data-sets for our experiments: KITTI 12 [6] and KITTI 15 [22], that we combined into a KITTI set, and Flying Things3D [20] summarized in Table 3.
Dataset Splits Yes We make validation sets by withholding 500 images from the Flying Things3D training set, and 58 from the KITTI training set, respectively.
Hardware Specification Yes We also acknowledge the support of NVIDIA Corporation with the donation of the Ge Force GTX TITAN X used for this research.
Software Dependencies No Our experiments are done with the Py Torch framework [26].
Experiment Setup Yes Table 2: Summary of training settings for every dataset. Flying Things3D KITTI Mode from scratch fine-tune Lr. schedule 0.01 for 120k, half every 20k 0.005 for 50k, half every 20k Iter. # 160k 100k Tr. image size 960 540 full-size 1164 330 Max disparity 255 255 Augmentation not used mix Up [42], anisotropic zoom, random crop