Dmytro Mishkin
1 min readJul 18, 2019

--

Don`t want to be too negative, but Dosovitskiy`s method doesn`t really work in practice. It even wasn`t properly tested.

The same for DeepHomography.

Deep detectors and descriptors which work in practice are: HardNet, SuperPoint, and D2Net: see very recent and relevant CVPR workshop
https://image-matching-workshop.github.io/

Regarding image sampling stuff: ASIFT and MODS.

Regarding Medical image registration: https://anhir.grand-challenge.org/Workshop-ISBI19/

--

--

Dmytro Mishkin
Dmytro Mishkin

Written by Dmytro Mishkin

Computer Vision researcher and consultant. Co-founder of Ukrainian Research group “Szkocka”.

Responses (1)