Dror Moran

[Dror Moran @DrorMoran]

@yuval_margalit_ @GuyTros @FadiKhatib1311 @GalunMeirav @RonenBasri [2/2] and models the displacement noise in inlier matches. Together with our two-step noise-aware training, our model achieves state-of-the-art results on both indoor and outdoor datasets while being faster than previous methods.

[Dror Moran @DrorMoran]

The code for our #neurips2024 paper "Consensus Learning with Deep Sets for Essential Matrix Estimation" is now online!

Paper: https://t.co/1H7AZYGu01
Code: https://t.co/mVfGJK24Pz
@yuval_margalit_ @GuyTros @FadiKhatib1311 @GalunMeirav @RonenBasri

[Dror Moran @DrorMoran]

@yuval_margalit_ @GuyTros @FadiKhatib1311 @GalunMeirav @RonenBasri [1/2] TL;DR: We propose a simple network architecture based on Deep Sets for estimating the relative position and orientation of two cameras. Given a collection of point matches extracted from two images, our method identifies outlier point matches

[Ronen Basri @RonenBasri]

Our exciting new paper applies spectral analysis to the neural tangent kernel to unravel the implicit bias of deep, over-parametrized *CNNs*, revealing advantages over fully-connected networks and underscoring the role of hierarchy. https://t.co/FcLch1sagX

[Yoni Kasten @yoni_kasten]

Our video lectures from the course Multiple View Geometry for Computer Vision Applications at Weizmann Institute of Science are now available online. https://t.co/SfcSNAoxYe
@RonenBasri @GalunMeirav @AmnonGeifman @HodayaKoslowsky

[Dror Moran @DrorMoran]

Check out the code for our ICCV 2021 oral presentation paper "Deep Permutation Equivariant Structure from Motion"!

@HodayaKoslowsky @yoni_kasten @HaggaiMaron @GalunMeirav @RonenBasri

https://t.co/FvPaPjFns1

[Dror Moran @DrorMoran]

Our architecture supports two setups: (1) single scene reconstruction and (2) learning from multiple scenes. Our experiments indicate that our method accurately recovers pose and structure, on par with state-of-the-art methods. Here are several examples: (cameras in red)

[Dror Moran @DrorMoran]

New paper: Deep Permutation Equivariant Structure from Motion.

We propose an architecture that, given a set of point tracks in multiple images of a scene, recovers both the camera params and 3D structure by minimizing an unsupervised reprojection loss

https://t.co/Rh83NGcOx5

[Dror Moran @DrorMoran]

Our network architecture is designed to respect the structure of the problem: the sought output is equivariant to permutations of both cameras and scene points.