Visual Inference Lab
LightNews LightNews
banner
visinf.bsky.social
Visual Inference Lab
@visinf.bsky.social
210 followers 330 following 42 posts
Visual Inference Lab of @stefanroth.bsky.social at @tuda.bsky.social - Research in Computer Vision and Machine Learning.

See https://www.visinf.tu-darmstadt.de/visual_inference
Posts Replies Media Videos
visinf.bsky.social
[7/8] Efficient Masked Attention Transformer for Few-Shot Classification and Segmentation

by @dustin-carrion.bsky.social, @stefanroth.bsky.social @simoneschaub.bsky.social 



🌍 visinf.github.io/emat
📄 arxiv.org/abs/2507.23642
💻 github.com/visinf/emat

Poster: Sun 4:40 PM - Exhibit Hall II #73
October 19, 2025 at 3:35 PM
visinf.bsky.social
[6/8] Motion-Refined DINOSAUR for Unsupervised Multi-Object Discovery (Oral at ILR+G Workshop)

by Xinrui Gong*, @olvrhhn.bsky.social *, @christophreich.bsky.social , Krishnakant Singh, @simoneschaub.bsky.social , @dcremers.bsky.social @stefanroth.bsky.social
October 19, 2025 at 3:35 PM
visinf.bsky.social
[5/8] ART: Adaptive Relation Tuning for Generalized Relation Prediction

by Gopika Sudhakaran, Hikaru Shindo, Patrick Schramowski, @simoneschaub.bsky.social, @kerstingaiml.bsky.social, @stefanroth.bsky.social

📄 arxiv.org/abs/2507.23543

Poster: Wed 2:45 PM, Exhibit Hall I #1501
October 19, 2025 at 3:35 PM
visinf.bsky.social
[4/8] Activation Subspaces for Out-of-Distribution Detection

by Baris Zongur, @robinhesse.bsky.social, @stefanroth.bsky.social

📄 arxiv.org/abs/2508.21695

Poster: Tue 11:45 AM, Exhibit Hall I #326
October 19, 2025 at 3:35 PM
visinf.bsky.social
[3/8] Feed-Forward SceneDINO for Unsupervised Semantic Scene Completion

by @jev-aleks.bsky.social *, @christophreich.bsky.social *, @fwimbauer.bsky.social , @olvrhhn.bsky.social , Christian Rupprecht, @stefanroth.bsky.social, @dcremers.bsky.social

🌍 visinf.github.io/scenedino/
October 19, 2025 at 3:35 PM
visinf.bsky.social
[2/8] Removing Cost Volumes from Optical Flow Estimators (Oral)

by @skiefhaber.de , @stefanroth.bsky.social @simoneschaub.bsky.social

🌍 visinf.github.io/recover

Talk: Tue 09:30 AM, Kalakaua Ballroom

Poster: Tue 11:45 AM, Exhibit Hall I #76
October 19, 2025 at 3:35 PM
visinf.bsky.social
TLDR: Given images captured by two vertically stacked omnidirectional cameras, DFI–OmniStereo integrates a pre-trained monocular depth foundation model into an iterative stereo matching approach. DFI–OmniStereo improves depth estimation accuracy, significantly outperforming the previous SotA.
October 17, 2025 at 9:27 PM
visinf.bsky.social
📢Excited to share our IROS 2025 paper “Boosting Omnidirectional Stereo Matching with a Pre-trained Depth Foundation Model”!

Work by Jannik Endres, @olvrhhn.bsky.social, Charles Cobière, @simoneschaub.bsky.social, @stefanroth.bsky.social and Alexandre Alahi.
October 17, 2025 at 9:27 PM
visinf.bsky.social
Activation Subspaces for Out-of-Distribution Detection (ICCV 2025)

by Baris Zongur, @robinhesse.bsky.social, and @stefanroth.bsky.social

📄: arxiv.org/abs/2508.21695

Talk: Wednesday, 1:30 PM, Oral Session 2
Poster: Wednesday, 3:30 PM, Poster 12
September 23, 2025 at 5:46 PM
visinf.bsky.social
Removing Cost Volumes from Optical Flow Estimators (ICCV 2025 Oral)

by @skiefhaber.de, @stefanroth.bsky.social, and @simoneschaub.bsky.social

🌍: visinf.github.io/recover

Poster: Friday, 10:30 AM, Poster 14
September 23, 2025 at 5:46 PM
visinf.bsky.social
Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 Highlight)

by @olvrhhn.bsky.social *, @christophreich.bsky.social *, @neekans.bsky.social, @dcremers.bsky.social, Christian Rupprecht, @stefanroth.bsky.social

🌍: visinf.github.io/cups

Poster: Thursday, 1:30 PM, Poster 19
September 23, 2025 at 5:46 PM
visinf.bsky.social
🌍: visinf.github.io/scenedino/

Talk: Friday, 10:00 AM, Oral Session 5
Poster: Friday, 10:30 AM, Poster 12
September 23, 2025 at 5:46 PM
visinf.bsky.social
Efficient Masked Attention Transformer for Few-Shot Classification and Segmentation (GCPR 2025)

by @dustin-carrion.bsky.social, @stefanroth.bsky.social, and @simoneschaub.bsky.social

🌍: visinf.github.io/emat

Poster: Wednesday, 03:30 PM, Postern 8
September 23, 2025 at 5:46 PM
visinf.bsky.social
Some impressions from our VISINF summer retreat at Lizumer Hütte in the Tirol Alps — including a hike up Geier Mountain and new research ideas at 2,857 m! 🇦🇹🏔️
August 29, 2025 at 12:48 PM
visinf.bsky.social
We had a great time at #CVPR2025 in Nashville!
June 20, 2025 at 5:04 AM
visinf.bsky.social
Disentangling Polysemantic Channels in Convolutional Neural Networks

by @robinhesse.bsky.social, Jonas Fischer, @simoneschaub.bsky.social, and @stefanroth.bsky.social

Paper: arxiv.org/abs/2504.12939

Talk: Thursday 11:40 AM, Grand ballroom C1
Poster: Thursday, 12:30 PM, ExHall D, Poster 31-60
June 11, 2025 at 8:56 PM
visinf.bsky.social
GLASS: Guided Latent Slot Diffusion for Object-Centric Learning

by Krishnakant Singh, @simoneschaub.bsky.social, and @stefanroth.bsky.social

Project Page: visinf.github.io/glass

Sunday 4:00 PM, ExHall D, Poster 239
June 11, 2025 at 8:56 PM
visinf.bsky.social
Scene-Centric Unsupervised Panoptic Segmentation

by @olvrhhn.bsky.social , @christophreich.bsky.social , @neekans.bsky.social , @dcremers.bsky.social, Christian Rupprecht, and @stefanroth.bsky.social

Sunday, 8:30 AM, ExHall D, Poster 330
Project Page: visinf.github.io/cups
June 11, 2025 at 8:56 PM
visinf.bsky.social
✅ Outperform the SotA by a significant margin.
✅ Generalize to different datasets, including an OOD setting.
✅ Stable performance across domains, different from supervised learning.
April 4, 2025 at 1:38 PM
visinf.bsky.social
We train a panoptic network on our high-precision pseudo-labels using self-enhanced copy-paste augmentation. Self-training refines predictions via a momentum network, aligning and filtering augmented outputs into self-labels. 🚀
April 4, 2025 at 1:38 PM
visinf.bsky.social
CUPS consists of pseudo-label generation, network bootstrapping, and self-training. We take inspiration from Gestalt principles (similarity, invariance, & common fate) and generate high-res. panoptic pseudo labels by complementing visual representations with depth & motion cues.
April 4, 2025 at 1:38 PM
visinf.bsky.social
📢 #CVPR2025 Highlight: Scene-Centric Unsupervised Panoptic Segmentation 🔥

We present CUPS, the first unsupervised panoptic segmentation method trained directly on scene-centric imagery.
Using self-supervised features, depth & motion, we achieve SotA results!

🌎 visinf.github.io/cups
April 4, 2025 at 1:38 PM
visinf.bsky.social
🏔️⛷️ Looking back on a fantastic week full of talks, research discussions, and skiing in the Austrian mountains!
January 31, 2025 at 7:38 PM
visinf.bsky.social
Want to learn about how model design choices affect the attribution quality of vision models? Visit our #NeurIPS2024 poster on Friday afternoon (East Exhibition Hall A-C #2910)!

Paper: arxiv.org/abs/2407.11910
Code: github.com/visinf/idsds
December 13, 2024 at 10:10 AM
visinf.bsky.social
Try PriMaPs with your own photos using our 🤗 demo at huggingface.co/spaces/olvrh...

@hf.co @gradio-hf.bsky.social
November 28, 2024 at 5:41 PM