Computer Vision and Machine Learning at MPI Informatics

People often use synthetic corruptions to test model robustness, but do these reflect real-world challenges?

We explore this in detail in our CVPR 2025 Workshop paper:

Are Synthetic Corruptions A Reliable Proxy For Real-World Corruptions?

arxiv.org/abs/2505.04835

by: @margretkeuper.bsky.social

🚨Deadline Extension Alert!

Our Non-proceedings track is open till August 15th for the eXCV workshop at ICCV.

Our nectar track accepts published papers, as is.

More info at: excv-workshop.github.io

@iccv.bsky.social #ICCV2025

⏳Still need to wait for your last experiment results?
📣 We're pleased to announce that the deadline for non-proceeding track #CV4DC at @iccv.bsky.social has been extended to August 15, 2025

Looking forward to your submissions! cv4dc.github.io/2025/

📄 spatialreasoners.github.io
🔗 github.com/spatialreaso...

Image for "🌀Spatial Reasoners for Continuous Variables in Any Domains"

4/ "🌀Spatial Reasoners for Continuous Variables in Any Domains" by @bartpog.bsky.social, @chriswewer.bsky.social, Bernt Schiele, and @janericlenssen.bsky.social (CODEML Workshop)

🔍 Software framework for training Spatial Reasoning Models in any domain

🔍 Can you really trust the explanations your classifier gives you? We show which pixels in the input are provably important to the classifier’s prediction within a radius around the input.

📄 openreview.net/pdf?id=NngoE...
🔗 github.com/AlaaAnani/ce...

Image for "Pixel-level Certified Explanations via Randomized Smoothing"

3/ "Pixel-level Certified Explanations via Randomized Smoothing" by @aanani.bsky.social, Tobias Lorenz, Mario Fritz, and Bernt Schiele

Spatial Reasoning with Denoising Models We introduce Spatial Reasoning Models (SRMs), a framework to perform reasoning over sets of continuous variables via denoising generative models. SRMs infer continuous representations on a set of unob...

📄 arxiv.org/abs/2502.21075
🔗 geometric-rl.mpi-inf.mpg.de/srm/
🔗 github.com/Chrixtar/SRM

Image for "Spatial Reasoning with Denoising Models"

2/ "Spatial Reasoning with Denoising Models" by @chriswewer.bsky.social, @bartpog.bsky.social, Bernt Schiele, and @janericlenssen.bsky.social

🔍 Can image generators solve visual Sudoku? Naively, no, with sequentialization and the correct order, they can!

DCBM: Data-Efficient Visual Concept Bottleneck Models Concept Bottleneck Models (CBMs) enhance the interpretability of neural networks by basing predictions on human-understandable concepts. However, current CBMs typically rely on concept sets extracted ...

📄 arxiv.org/abs/2412.11576
🔗 github.com/KathPra/DCBM

Image for "DCBM: Data-Efficient Visual Concept Bottleneck Models"

1/ "DCBM: Data-Efficient Visual Concept Bottleneck Models" by @katharinaprasse.bsky.social*, @patrickknab.bsky.social*, Sascha Marton, Christian Bartelt, and @margretkeuper.bsky.social

🔍 Data-efficient CBMs (DCBMs) generate concepts from image regions detected by segmentation or detection models

Papers accepted at ICML 2025 from the Computer Vision and Machine Learning Department at the Max Planck Institute for Informatics.

Papers being presented from our group at #ICML2025!

Congratulations to all the authors! To know more, visit us in the poster sessions!

A 🧵with more details:

@icmlconf.bsky.social @mpi-inf.mpg.de

📣 Proceeding track's results are out.
🎉 Congratulations to all the authors whose papers were accepted. We can't wait to meet you at @iccv.bsky.social in Hawaii on Oct 19th.

⏰ Our non-proceeding track is still accepting submissions until July 20th! Details in the comments

🎉 Congrats to Yue Fan on defending his PhD: "Improving Representation Learning from Data and Model Perspectives: Semi-Supervised Learning and Foundation Models" 🧑‍🎓

He is now at Genmo.ai as a Research Engineer working on video generation! 🚀

More: yue-fan.github.io

All the best!

Congratulations to our PhD alumna @annakukleva.bsky.social for being awarded the prestigious Otto Hahn Medal by @maxplanck.de! 🎉

🚀 Just accepted to ICCV 2025!

In DIY-SC, we improve foundational features using a light-weight adapter trained with carefully filtered and refined pseudo-labels.

🔧 Drop-in alternative to plain DINOv2 features!
📦 Code + pre-trained weights available now.
🔥 Try it in your next vision project!

Submission deadline is today!

Submission Deadline is extended by 6 days.

#ICCV2025 @iccv.bsky.social

@mattiasegu.bsky.social

A heart congratulations to the freshly minted Dr. Mattia Segù on successfully defending his PhD, Congratulazioni!!! 🎉 🎓.

His thesis is titled: Learning to Track: From Limited Supervision to Long-range Sequence Modeling

Checkout his web-page to learn more about his work: mattiasegu.github.io

Call for papers at the eXCV workshop at ICCV 2025.

Join us in taking stock of the state of the field of explainability in computer vision, at our Workshop on Explainable Computer Vision: Quo Vadis? at #ICCV2025!

@iccv.bsky.social

At #CVPR2025 and working on consistency in video and multi-view generative models?

Come and visit our poster on Friday afternoon, where I present 𝗠𝗘𝘁𝟯𝗥: 𝗠𝗲𝗮𝘀𝘂𝗿𝗶𝗻𝗴 𝗠𝘂𝗹𝘁𝗶-𝗩𝗶𝗲𝘄 𝗖𝗼𝗻𝘀𝗶𝘀𝘁𝗲𝗻𝗰𝘆 𝗶𝗻 𝗚𝗲𝗻𝗲𝗿𝗮𝘁𝗲𝗱 𝗜𝗺𝗮𝗴𝗲𝘀

@mohammadasim98.bsky.social @wimmerthomas.bsky.social @mpi-inf.mpg.de @cvml.mpi-inf.mpg.de

7 / 🧵 ...
Workshop: Women in Computer Vision (WiCV)
📱 @sukrutrao.bsky.social @SwetaMahajan @MoritzBoehle

7/ 🧵 Discover-then-Name: Task-Agnostic Concept Bottlenecks via Automated Concept Discovery
Authors: S. Rao, S. Mahajan, M. Böhle, B. Schiele
🔍 Explore sparse autoencoders to automatically extract and name concepts, enabling performance improvements on downstream tasks.
📚 arxiv.org/abs/2407.14499

6 / 🧵 ...
Workshop: Women in Computer Vision (WiCV)
📱 @tejaswinimedi.bsky.social @margretkeuper.bsky.social

6/ 🧵 3D-WAG: Wavelet-Guided Autoregressive Generation for 3D Shapes
Authors: T. Medi*, A. Rampini, P. Reddy, P. K. Jayaraman, M. Keuper
🔍 3D-WAG introduces wavelet-guided autoregressive generation for 3D shapes, aiming for better geometry modeling.
📚 arxiv.org/abs/2411.19037

5/ 🧵 ...
Workshop: Explainable AI for Computer Vision (XAI4CV)
📱 @katharinaprasse.bsky.social @smarton.bsky.social @margretkeuper.bsky.social

5/ 🧵 Data-Efficient Visual Concept Bottleneck Models
Authors: K. Prasse, P. Knab, S. Marton, C. Bartelt, M. Keuper
🔍 Introducing data-efficient visual concept bottleneck models for improved explainability in CV.
📚 arxiv.org/abs/2412.11576

4/ 🧵 ...
Workshops: What is Next in Multimodal Foundation Models? | Women in Computer Vision
📱 @maheensaleh.bsky.social @ninashv.bsky.social @annakukleva.bsky.social @hildekuehne.bsky.social

4/ 🧵 HD-VILA-Caption: A Diverse Video-Text Dataset Derived from ASR Narrations
By: M. Saleh, N. Shvetsova, A. Kukleva, H. Kuehne, B. Schiele
🔍 HD-VILA-Caption is a large-scale, diverse video-text dataset with 10M high-quality captions, built from ASR subtitles for video-language pretraining.