Arjun Karpur

Head of ML @ Epsilon Health | Prev. @GoogleDeepMind @UTCompSci | Training image encoders & VLMs for radiology

[Arjun Karpur @arjunkarpur]

Epsilon is hiring, come join us!

We're looking for Research Scientists, Research Engineers, and Full Stack SWEs to help us scale vision encoders and VLMs for radiology🩻

Jobs: https://t.co/Vjf6tuNe3G

Website: https://t.co/tEA7o0NDQq

Or email me! arjun [at] epsilonlabs . ai

[Arjun Karpur @arjunkarpur]

And I’m also excited to support my old teammates as we share our paper:

TIPSv2: Advancing Vision-Language Pretraining with Enhanced Patch-Text Alignment

https://t.co/rcm53KnutX

^ please stop our poster on Sun to learn about training vision encoders w/ patch-level text info

[Arjun Karpur @arjunkarpur]

On the way to Denver for #CVPR2026! 🛫

Looking forward to meeting folks working in the multimodal / visual grounding / medical imaging space, and to reconnecting with old friends!

Let me know if you’ll be there and want to meet up

[André Araujo @andrefaraujo]

True multimodal AI needs to understand the world spatially 🎯
🚀 Excited to release #CVPR2026 TIPSv2 from @GoogleDeepMind, a foundational image-text encoder with spatial awareness, leading to strong overall results and massive gains on patch-text alignment. 🔥
1/N

[Arjun Karpur @arjunkarpur]

Thanks to everyone that stopped by today! Open-source code and weights for TIPS are available at:

https://t.co/PqdAZYETKu

#ICLR2025

[Arjun Karpur @arjunkarpur]

Excited to be presenting TIPS at this morning’s #ICLR2025 poster session! Come by poster #318 and say hi 👋

w/ @kfrancischen @andrefaraujo @kmaninis #ICLR #ICLR25

[Arjun Karpur @arjunkarpur]

@StanSzymanowicz Congrats on the release, Stan!

[André Araujo @andrefaraujo]

Multimodal AI encoders often lack spatial understanding… but not anymore! Our #ICLR2025 TIPS model (Text-Image Pretraining with Spatial awareness) from @GoogleDeepMind can help 💡🚀
Check out our strong & versatile image-text encoder 💪
Paper & code: https://t.co/LCiqV4gaQ0

[Kevis-Kokitsi Maninis @kmaninis]

📢📢 We released checkpoints and Pytorch/Jax code for TIPS: https://t.co/0JUIRML8gr

Paper updated with distilled models, and more:
https://t.co/zebYMD0VFz

#ICLR2025

[André Araujo @andrefaraujo]

Excited to release a super capable family of image-text models from our TIPS #ICLR2025 paper! https://t.co/1scX7H1DIb

We have models from ViT-S to -g, with spatial awareness, suitable to many multimodal AI applications. Can’t wait to see what the community will build with them!

[André Araujo @andrefaraujo]

Great work from our team at @GoogleDeepMind!
With @kmaninis, @kfrancischen, @sohamg121, @arjunkarpur, Koert Chen, Ye Xia, Bingyi Cao, @GuangxingHan, Jan Dlabal, Dan Gnanapragasam, Mojtaba Seyedhosseini, @howardzzh

[Arjun Karpur @arjunkarpur]

OmniGlue on display at #CVPR24

Thanks for presenting our work! @hanwenjiang1 @andrefaraujo

Full paper: https://t.co/gMjqthXxNR

[Arjun Karpur @arjunkarpur]

@AlphaRealcat I really love the image matching webui, glad OmniGlue is now an option there. Thank you!

[Vincent Qin @AlphaRealcat]

Just released an ONNX version of OmniGlue. No more need for TensorFlow installations, folks! 😄 Check out the comparison between the TensorFlow and ONNX versions in the image below. @arjunkarpur
Code: https://t.co/1skwudSlEz

[André Araujo @andrefaraujo]

Meet #CVPR2024 OmniGlue, the first learnable matcher designed with generalization as a core principle! Great performance on many domains, ideal for in-the-wild matching 🎯

Code available!

https://t.co/Kj0SHMQJRn

with @hanwenjiang1, @arjunkarpur, Bingyi Cao, @qixing_huang

[#CVPR2026 @CVPR]

HUGE shoutout to our #CVPR2024 Outstanding Reviewers 🫡

[Arjun Karpur @arjunkarpur]

@AlphaRealcat Thanks for the tweet! Adding OmniGlue to the image matching webui is on our TODO list

[Arjun Karpur @arjunkarpur]

@qixing_huang @hanwenjiang1 Thanks for the shoutout Qixing! I’m glad we had a chance to work together again

[Kevis-Kokitsi Maninis @kmaninis]

Are you evaluating 3D reconstruction/dense correspondences on synthetic datasets because real datasets are "not accurate enough"? Check out NAVI, a dataset that offers near-perfect alignments of 3D shapes on real image collections: https://t.co/zb68qaRaw8 #NeurIPS2023 (1/2)

[Zhenjun Zhao @zhenjun_zhao]

LFM-3D: Learnable Feature Matching Across Wide Baselines Using 3D Signals

Arjun Karpur, Guilherme Perrotta, Ricardo Martin-Brualla, Howard Zhou, Andre Araujo

tl;dr: normalized object coordinates/monocular depth estimates+improved PE->improve SuperGlue

https://t.co/QB6UpfefRy