Lukas Hoyer

[Computer Vision and Machine Learning - MPI-INF @cvml_mpiinf]

"Scribbles for All: Benchmarking Scribble Supervised Segmentation Across Datasets" by Wolfgang Boettcher, @lukashoyer3, @uenalozan, @janericlenssen, and Bernt Schiele.

Wed 11 Dec 4:30 p.m. PST–7:30 p.m. PST, East Exhibit Hall A-C #1702

[Ferjad Naeem @ferjadnaeem]

We are presenting SILC in the after noon session at #ECCV2024.
Stop by if you want the best CLIP model for dense prediction tasks :)

SILC: Improving Vision Language Pretraining with Self-Distillation
🗓 Thu 3 Oct 4:30 p.m. — 6:30 p.m. CEST Poster 204
💻 https://t.co/sW9kMhVpCo

[Anton Obukhov @AntonObukhov1]

Woohoo, DGInStyle is accepted at ECCV 2024! I am looking forward to meeting many of you in Milan! https://t.co/s6xBeoetGc

[Anton Obukhov @AntonObukhov1]

Another gem from our lab — DGInStyle! We use Stable Diffusion to generate semantic segmentation data for autonomous driving and train domain-generalizable networks.

📟 Website: https://t.co/B0a7mv14gq
🧾 Paper: https://t.co/BlMHsJLlzl
🤗 Model: https://t.co/M0xRT8mHPc
🐙 Code: https://t.co/W8uuJ7znyl

Team: Yuru Jia, Lukas Hoyer (@lukashoyer3), Shengyu Huang (@ShengyHuang), Tianfu Wang (@TianfuWang2),
Luc Van Gool, Konrad Schindler, and Anton Obukhov (@AntonObukhov1).

[Lukas Hoyer @lukashoyer3]

@ducha_aiki We only compared ImageNet and CLIP pre-training. Here, CLIP pre-training works much better due to the rich semantic prior. We did not benchmark DINOv2 as CLIP comes with the conceptual advantage of a semantic alignment between text (i.e. class name) and image embeddings.

[Lukas Hoyer @lukashoyer3]

🚀Introducing SemiVL - a semi-supervised semantic segmentation method exploiting priors from vision-language models

📄Paper: https://t.co/bEOBn17vD5
💻Code: https://t.co/4qnEHv2uZb
😊Team: @lukashoyer3, David J. Tan, @ferjadnaeem, Luc Van Gool, @fedassa

#ETHZurich #Google

[Lukas Hoyer @lukashoyer3]

Introducing SemiVL - a semi-supervised semantic segmentation model exploiting priors from vision-language models.

Paper: https://t.co/DlRhRjnIqG
Code: https://t.co/4qnEHv1X9D
Team: @lukashoyer3, David J. Tan, @ferjadnaeem, Luc Van Gool, @fedassa

#ETHZurich #Google internship

[Lukas Hoyer @lukashoyer3]

We extend our previous DAFormer and HRDA beyond synthetic-to-real adaptation. We show their capabilities for day-to-night and clear-to-adverse-weather adaptation and extend them to domain generalization on unseen domains. We improve the SOTA by more than 10 mIoU on 5 benchmarks.

[Lukas Hoyer @lukashoyer3]

I'm happy to announce that our paper "Domain Adaptive and Generalizable Network Architectures and Training Strategies for Semantic Image Segmentation" was accepted by TPAMI!

https://t.co/Y5SFB4x2Ii
https://t.co/yd8ZBLffGK

[Lukas Hoyer @lukashoyer3]

I'm happy to share that our paper "EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation" was accepted at #ICCV2023!

[Suman Saha @SumanSa51694405]

We are delighted to share that our paper "EDAPS: Enhanced Domain-Adaptive Panoptic Segmentation" by Suman Saha, Lukas Hoyer, Anton Obukhov, Dengxin Dai, and Luc Van Gool was accepted at #ICCV2023!
arXiv: https://t.co/ryGlqLhz2e
@lukashoyer96 @AntonObukhov1

[Lukas Hoyer @lukashoyer3]

I will present "MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation" at #CVPR23 booth 333 this morning. If you are interested, come by or check out our video presentation https://t.co/F5fsEdLjZV!

[Lukas Hoyer @lukashoyer3]

We are happy to announce that our paper "MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation" was accepted at #CVPR23!
By enforcing consistency of predictions from masked images, MIC enhances the adaptation to an unlabeled target domain. https://t.co/IlYzf9so4R

[Parallel Domain @parallel_domain]

@lukashoyer96 speaking at the #CVPR2023 SDAS workshop on Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation

[Lukas Hoyer @lukashoyer3]

We are happy to announce that our paper "Improving Semi-Supervised and Domain-Adaptive Semantic Segmentation with Self-Supervised Depth Estimation" was accepted at #IJCV!

Open Access Paper: https://t.co/TrxG8LNttn
GitHub: https://t.co/FxLnBdALin

[Lukas Hoyer @lukashoyer3]

We are happy to announce that our paper "MIC: Masked Image Consistency for Context-Enhanced Domain Adaptation" was accepted at #CVPR23!
By enforcing consistency of predictions from masked images, MIC enhances the adaptation to an unlabeled target domain. https://t.co/IlYzf9so4R

[Lukas Hoyer @lukashoyer3]

I will present our paper "HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation" at #ECCV2022 in Tel Aviv tomorrow (25.10.) from 15:30 to 17:30. Feel free to come by for interesting discussions or check out our video https://t.co/aNoFEBqWrt.

[Lukas Hoyer @lukashoyer3]

We are happy to announce that our paper "HRDA: Context-Aware High-Resolution Domain-Adaptive Semantic Segmentation" was accepted at #ECCV2022!
HRDA enables adapting small objects and preserving fine segmentation details as shown in the video.
Paper&Code: https://t.co/yd8ZBLffGK

[Lukas Hoyer @lukashoyer3]

I just received the physical ETH Medal for my Master's thesis. It has even the name engraved! I would like to thank Dengxin Dai and Luc Van Gool for their great supervision, which made this project possible. If you are interested, feel free to visit https://t.co/FxLnBdBj7V

[Lukas Hoyer @lukashoyer3]

I will present our DAFormer at #CVPR2022 today from 14:00 to 17:00 at the poster booth 193. Feel free to also check out our video https://t.co/t7gI3OCPsT and come by for interesting discussions.

[AK @_akhaliq]

DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation
abs: https://t.co/IePs79GeHI
github: https://t.co/sI0P4olATR

improves the sota performance by 10.8 mIoU for GTA→Cityscapes and 5.4 mIoU for Synthia→Cityscapes

[Lukas Hoyer @lukashoyer3]

@followML_ Thanks a lot for sharing our paper, @followML_! Let me complement the title page with some visual examples.

[Lukas Hoyer @lukashoyer3]

We are happy to announce that our paper "DAFormer: Improving Network Architectures and Training Strategies for Domain-Adaptive Semantic Segmentation" was accepted at #CVPR2022. It improves the SOTA by 19% on GTA->Cityscapes.
https://t.co/SPwQD08fz3
https://t.co/d8jDLMqrFM