#VisualOverload
Is basic image understanding solved in today’s SOTA VLMs? Not quite.

We present VisualOverload, a VQA benchmark testing simple vision skills (like counting & OCR) in dense scenes. Even the best model (o3) only scores 19.8% on our hardest split.
September 8, 2025 at 3:28 PM
August 4, 2026 at 3:01 PM
Proud to announce that VisualOverload was accepted to #CVPR2026! Overall 2/2 accepted.
🚨 New paper out!
"VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes"
👉 arxiv.org/abs/2509.25339
We test 37 VLMs on 2,700+ VQA questions about dense scenes.
Findings: even top models fumble badly—<20% on the hardest split and key failure modes in counting, OCR & consistency.
February 21, 2026 at 3:14 PM
🚨 New paper out!
"VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes"
👉 arxiv.org/abs/2509.25339
We test 37 VLMs on 2,700+ VQA questions about dense scenes.
Findings: even top models fumble badly—<20% on the hardest split and key failure modes in counting, OCR & consistency.
October 1, 2025 at 1:17 PM
Do Vision-Language Models (VLMs) actually "see" everything in a crowded room? 🔍

Today at #CVPR2026, we are presenting VisualOverload, our work exploring the critical visual perception bottlenecks of VLMs in dense scenes.

📍 Today (Poster Session 6), 5:30 PM - 7:30 PM, Poster 431 (ExHall A)
June 7, 2026 at 5:14 PM
Chaos has a face. Actually… dozens of them. And they're all screaming. Welcome to the cartoon apocalypse.

#SurrealMayhem #CartoonChaos #AIArt #ScreamingFaces #AbsurdistArt #SkullsAndGiggles #VisualOverload #WhimsicalTerror #Surrealism #BlueskyArt #artspace.ai
May 12, 2025 at 10:18 PM
September 17, 2025 at 10:11 PM
🌎 paulgavrikov.github.io/visualoverload

Joint work with Wei Lin, M. Jehanzeb Mirza, Soumya Jahagirdar, Muhammad Huzaifa, Sivan Doveh, Serena Yeung-Levy, James Glass, Hilde Kuehne.
VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes
The paper introduces VisualOverload, a new visual question answering (VQA) benchmark designed to test vision-language models (VLMs) on densely populated, detail-rich scenes using public-domain paintin...
paulgavrikov.github.io
June 7, 2026 at 5:14 PM
📊 VisualOverload =
• 2,720 Q–A pairs
• 6 vision tasks
• 150 fresh, high-res, royalty-free artworks
• Privately held ground-truth responses
September 8, 2025 at 3:28 PM
Gavrikov, Lin, Mirza, Jahagirdar, Huzaifa, Doveh, Yeung-Levy, Glass, Kuehne: VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes https://arxiv.org/abs/2509.25339 https://arxiv.org/pdf/2509.25339 https://arxiv.org/html/2509.25339
October 1, 2025 at 6:30 AM
VisualOverload, a new benchmark of 2,720 Q&A pairs from high‑resolution public‑domain paintings, evaluated 37 VLMs; top model o3 reached 69.5% overall but only 19.6% on the hardest split. https://getnews.me/visualoverload-benchmark-shows-gaps-in-vision-language-models/ #visualoverload #benchmark
October 3, 2025 at 11:50 AM
Paul Gavrikov, Wei Lin, M. Jehanzeb Mirza, Soumya Jahagirdar, Muhammad Huzaifa, Sivan Doveh, Serena Yeung-Levy, James Glass, Hilde Kuehne
VisualOverload: Probing Visual Understanding of VLMs in Really Dense Scenes
https://arxiv.org/abs/2509.25339
October 1, 2025 at 11:15 AM
THE MESS YOU NORMALISED

At first, it bothered you.
Then you adapted to it.

Piles on desks.
Boxes in corners.
Things everywhere.

Humans adapt fast.
Even to chaos.

#ChaosBecomesNormal #VisualOverload #DeclutterAwareness #MentalEnvironment #jrmendoza
May 13, 2026 at 4:00 PM