iiinova — where your digital self keeps evolving

Sonat Baltacı

@sonatbaltaci.bsky.social

102 followers 104 following 7 posts

PhD student in computer vision at Imagine, ENPC.

Posts Replies Media

Sonat Baltacı @sonatbaltaci.bsky.social · 20/07/2026

Joint work with @rloiseau.bsky.social & M. Aubry! 🌐 Webpage & findings: imagine.enpc.fr/~sonat.balta... 💻 Code: github.com/sonatbaltaci... 📄 ArXiv: arxiv.org/abs/2604.19480 3/3

040

Sonat Baltacı @sonatbaltaci.bsky.social · 20/07/2026

We dive into their design mechanics, identify core components, and perform an extensive analysis on clustering benchmarks. Leveraging these findings, we propose a new sprite-based method that scales linearly with the number of objects and identifies object categories in an interpretable way. 2/3

140

Sonat Baltacı @sonatbaltaci.bsky.social · 20/07/2026

Excited to share that our paper on sprite-based image decomposition is accepted at TMLR! 🎉 Sprite-based models are highly interpretable but struggle to scale to complex, multi-object images.   1/3

1204

Sonat Baltacı @sonatbaltaci.bsky.social · 24/06/2026

Joint work with @raphaelbaena.bsky.social, F. Meng, S. Norindr, F. Somer, M. Husson, and M. Aubry! 🌐 Webpage: imagine.enpc.fr/~sonat.balta... 💻 Code: github.com/sonatbaltaci... 📄 ArXiv: arxiv.org/abs/2606.15886 🤗 Dataset: huggingface.co/datasets/son...  4/4

060

Sonat Baltacı @sonatbaltaci.bsky.social · 24/06/2026

Detailed annotations: Ordered polygons capturing text boundaries and reading directions, plus 20 class labels for Latin diagrams.  Introducing Poly-DETR: a simple DINO-DETR extension designed to predict ordered polygon vertices. 3/4

140

Sonat Baltacı @sonatbaltaci.bsky.social · 24/06/2026

Highlights:  948 historical astronomical diagrams spanning 10 centuries (8th to 18th). 10,940 oriented polygonal text regions across 7 major linguistic traditions (Arabic, Persian, Chinese, Byzantine, Latin, Hebrew, and Sanskrit). 2/4

140

Sonat Baltacı @sonatbaltaci.bsky.social · 24/06/2026

Happy to share that our work “Text Region Detection in Historical Astronomical Diagrams” is accepted to #ICDAR2026! 🎉 An open-access dataset of 948 historical astronomical diagrams spanning 10 centuries and 7 linguistic traditions to tackle text detection in complex and diverse settings. 🌌 1/4

1237

Reposted by Sonat Baltacı

David Picard @davidpicard.eurosky.social · 08/04/2026

🚨 arxiv.org/abs/2604.06129 PoM: A Linear-Time Replacement for Attention with the Polynomial Mixer This paper is the result of doing a lab-wide hackathon on an idea I've had for some time. Probably the paper with the highest number of authors I've ever done. It's a CVPR Findings 26. Thread 🧵👇

arxiv.org

PoM: A Linear-Time Replacement for Attention with the Polynomial Mixer

This paper introduces the Polynomial Mixer (PoM), a novel token mixing mechanism with linear complexity that serves as a drop-in replacement for self-attention. PoM aggregates input tokens into a comp...

46520