Edgar
@edgarriba
Creator of @kornia_foss and co-founder of https://t.co/mYuSg1ClqG | Computer Vision Researcher
ID:249701367
https://github.com/edgarriba 09-02-2011 16:06:48
1,8K Tweets
1,6K Followers
1,2K Following
0.7.2 is out!
- Added DeDoDe features (thanks Johan Edstedt )
- LightGlue models, available nowhere else - DeDoDe (B/G), KeyNet-HardNet
- KMeans implementation
- New augmentations: RandomGaussianIllumination, RandomLinearIllumination, RandomLinearCorner
1/2
github.com/kornia/kornia/…
Multilinear Operator Networks
Yixin Cheng, Grigoris Chrysos, Markos Georgopoulos, Volkan Cevher
tl;dr: element-wise multiplication and layer norm is all you need (pun intended)
arxiv.org/abs/2401.17992
Grandmaster-Level Chess Without Search
Anian Ruoss, Grégoire Delétang, Sourabh Medapati, Jordi Grau-Moya, Kevin Li, Elliot Catt, John Reid, Tim Genewein
tl;dr: supervised learning labeled by Stockfish FTW.
Many ablations, also scaling experiment
arxiv.org/abs/2402.04494…
CodeIt: Self-Improving Language Models with Prioritized Hindsight Replay
Natasha Butt, Blaze(j) Manczak 🇵🇱🇱🇺🇪🇺, Auke Wiggers, Corrado Rainone, David Zhang, Michaël Defferrard, Taco Cohen
tl;dr: sample a program, try it, add to the replay pool.
New sota on ARC
arxiv.org/abs/2402.04858…
Deconstructing Denoising Diffusion Models for Self-Supervised Learning
Xinlei Chen, Zhuang Liu Saining Xie Kaiming He
tl;dr: adding the noise in the low-dim space (even PCA works) is the most important of diffusion models for representation learning.
arxiv.org/abs/2401.14404
AGILE3D: Attention Guided Interactive Multi-object 3D Segmentation
Yuanwen Yue Sabarinath Mahadevan Jonas Schult Francis Engelmann Bastian Leibe Konrad Schindler TheodoraKontogianni
tl;dr:if you are doing interactive segmentation - pre-extract features
#ICLR2024
arxiv.org/abs/2306.00977…
Kudos to Philipp Lindenberger and Paul-Edouard Sarlin for adding DoGHardNet model to official LightGlue trained by Kornia team.
That is a significant upgrade, if you have to use DoG(SIFT) detector.
As simple as
matcher = LightGlue(features='doghardnet').eval()
