Just published fastpose - a numba port of PoseLib with some changes. Same accuracy, but significantly faster: 2.9–9.7× faster on one CPU, 28–126× on GPU. Includes relpose, RePoseD, F, H, P3P and P4Pf.
Relpose benchmarks here:
kocurvik.github.io/fastpose
1
2
39
@Parskatt you may find this interesting. 10k iters with 16k RoMa v2 matches in 30 ms on A100, though in the benchmark using fewer matches led to greater accuracy.
1
30
Viktor Kocur retweeted
DGSfM: Depth-Guided Scale-Aware Global Structure-from-Motion
Sithu Aung, @kocurvik, Yaqing Ding, @SattlerTorsten, @ZKukelova
tl;dr: depth improves GLOMAP in relative pose estimation, view-graph and correspondence filtering, scale averaging, pose-point initialization
arxiv.org/abs/2607.09507
Viktor Kocur retweeted
DGSfM: Depth-Guided Scale-Aware Global Structure-from-Motion
Sithu Aung @kocurvik Yaqing Ding @SattlerTorsten @ZKukelova
tl;dr: carefully engineered pipeline with dopplegangers++ graph filtering, monodepth and other steps to help GloMAP. Good IMC21 eval
arxiv.org/abs/2607.09507
Check out our latest work at #ICCV2025 "RePoseD: Efficient Relative Pose Estimation With Known Depth Information" Oral Session 4B🕑2 p.m. (Hawaii Time) and poster #151.
tldr: We show how MDEs and relative pose can improve each other.
Code + demos here: kocurvik.github.io/reposed/
98
Viktor Kocur retweeted
RePoseD: Efficient Relative Pose Estimation With Known Depth Information
Yaqing Ding, @kocurvik, Václav Vávra, Zuzana Berger Haladová, Jian Yang, @SattlerTorsten, @ZKukelova
tl;dr: how MDE helps two-view relative pose estimation; comprehensive settings and MDE methods
arxiv.org/abs/2501.07742
kocurvik.github.io/reposed/
Viktor Kocur retweeted
check out this interesting workshop 🧐👉 The Role of Embodiment In Self-Aware Humans and Robots!
more info👇
#terais #horizoneurope #horizonproject
One week till the Workshop The Role of Embodiment In Self-Aware Humans and Robots!
Check out program contact.iit.it/tsc-2023-work… and if you already have questions about self-aware robots to our speakers you can write us down here contact.iit.it/tsc-2023-work…
Recording will be released😃
Recent paper from my postdoc at @FIT_VUT in collaboration with @NenoVision - Correction of artifacts and noise in atomic force microscopy images using deep learning - trained with synthetic data only authors.elsevier.com/a/1gM3g… #deeplearning #afm #imageprocessing
Viktor Kocur retweeted
Share of global CO₂ emissions vs. share of global population
The US – only 4.4% of the world population – emit 15% of global CO₂.
China – 19% of the world population – emit 29% of global CO₂.
From @HannahRitchie02’s latest post:
ourworldindata.org/share-co2…



















