Constantin Seibold

140 posts

Constantin Seibold

@cmseibold

Postdoctoral Researcher, UK-Essen

Essen Katılım Kasım 2020

251 Takip Edilen68 Takipçiler

Constantin Seibold retweetledi

Thilo Zimmermann (@thilozimmermann@bsky.social)

Thilo Zimmermann (@[email protected])@ThiloZimmermann·15 Tem

Neues KI-Institut @Uni_Stuttgart umfasst 2 @Cyber_Valley Prof. Analytic Computing (@ststaab) & #AutonomeSysteme (#KaiArras), @SimTechStuttga2 Prof. ML in den Sim.wiss., Jun.-Prof. Autonomous Sensing and Perception Systems (@AlinaRoitberg) & Prof. #MaschinellesLernen & #Robotik

Uni Stuttgart@Uni_Stuttgart

KI für #Simulationen und #IntelligenteRobotik: Daran forscht künftig das neugegründete Institut für Künstliche Intelligenz an der #UniStuttgart. Das Ziel: den Transfer von KI-Anwendungen in Wirtschaft und Gesellschaft vorantreiben.👉 bit.ly/3PSrwAd

Deutsch

1.1K

Constantin Seibold retweetledi

Vivek Sharma@vivoutlaw·17 Haz

Come join our #CVPR23 workshop on Large Scale Holistic Video Understanding (@LSHVU), happening on 18th starting at 8:30am PST. Program schedule (Room: East 8, Poster: West Exhibit Hall, # 100-114): lnkd.in/e2dzJ-AG

English

1.1K

Constantin Seibold@cmseibold·8 Haz

This was my last paper during my time at the graduate school @hidss4health and a collaboration with @AlexanderJaus, Matthias Fink, Moon Kim, @SimaelR, Ken Herrmann, @JensKleesiek and Rainer Stiefelhagen.

English

138

Constantin Seibold@cmseibold·8 Haz

We make this available via a simple "pip install cxas". We also provide a gradio demo to easily test this method: huggingface.co/spaces/cmseibo… For more details, check out our paper: arxiv.org/abs/2306.03934

English

158

Constantin Seibold@cmseibold·8 Haz

The anatomy in chest radiographs (CXR) is typically difficult segment manually due to the ambiguity introduced by overlapping structures. We Adress this in our paper and publish models to segment 157 anatomical labels in CXR: arxiv.org/abs/2306.03934

English

911

Constantin Seibold@cmseibold·23 May

When your otherwise free work day is suddenly filled with meetings... 😅

University of Oxford@UniofOxford

🐢 | During a research trip, a team of Oxford researchers took part in the #ProtetugaProject through @FundaPrincipe to observe sea turtles laying eggs. Fundação Príncipe works with local communities to protect biodiversity on the island. 📽 | @principeproject #TurtleDay

English

126

Constantin Seibold@cmseibold·21 Nis

Also a big thank you to the @hidss4health graduate school for enabling these past few years in research!

English

100

Constantin Seibold@cmseibold·21 Nis

I guess it is finally done!! Managed to successfully defend my PhD thesis today. I researched methods aiding the generation of medical reports. I loved my time at the cv:hci and the @KITKarlsruhe overall. I will certainly miss the atmosphere, great colleagues and friends!!

English

397

Constantin Seibold@cmseibold·20 Nis

Probably not thinking too straight right now due to an exam in the afternoon, but I think I might release a pre-print with associated Models and Dataset for fine-grained anatomy segmentation in the first week of May... (because self-set deadlines always work ... 😆)

English

293

Constantin Seibold@cmseibold·3 Nis

@ylecun Maybe I got this wrong but haven't masked autoencoders applied on vision transformers shown quite promising results? Do they not count toward generative models?

English

Yann LeCun@ylecun·3 Nis

Well, but learning image representations with generative models never actually worked. When it comes to SSL for images, GAN, VAE, denoising AE and other gen models have been a bust. What has worked is Joint Embedding (non generative) Architectures like Siamese nets, JEPA & such.

Sander Dieleman@sedielem

Feels like just yesterday that any work on generative modelling had to be justified through "representation learning", or some sort of downstream task, to be taken seriously. Times have changed!

English

186

117.9K

Constantin Seibold@cmseibold·14 Mar

Quite funny how such a notification can improve the mood 😄 scholar.google.com/citations?user…

English

205

Constantin Seibold@cmseibold·22 Oca

😓🖨️

QME

142

Constantin Seibold@cmseibold·26 Ara

@MariamAboian Do you mean by potentially highlighting differences to an expected healthy heart in this case?

English

Mariam Aboian@MariamAboian·26 Ara

@cmseibold When it comes to annotating images with pathology, it is critical to understand pathophysiology principles behind the disease. Annotations guided by biological principles will work better on clinical data. I dont have a paper to prove this…would be nice to actually show that 😅

English

Constantin Seibold@cmseibold·26 Ara

I've been going over the saliency maps of our CXR Vision-Language model from MICCAI22. Despite it being somewhat obvious, I found it interesting that for some classes it will just activate on the abnormal enlargements for cardiomegaly. Other CNN had activated on the entire heart.

English

561

Constantin Seibold@cmseibold·24 Ara

Merry Christmas to all of you! I hope we can get some time to rest and then freshly start the new year.

English

121

Constantin Seibold@cmseibold·11 Ara

The more you can't have something, the more you want it, right? Currently writing my thesis and so many ideas pop into my head of what I would love to just try out, but I don't have the timeeee... Feels like torture... 😭😭😭

English

Constantin Seibold@cmseibold·21 Kas

@iamzhouh There should have been a mail by Andrew Gilbert via eventbrite with a link afaik. Don't know any other way

English

Hurry@iamzhouh·21 Kas

@cmseibold How to enter this interface?

English

Constantin Seibold@cmseibold·21 Kas

Standing here in poster room 1 to present my paper about detailed chest Xray annotations for phrase grounding 🤸 Still wish I could have attended #BMVC2022 in person =(

English

Keşfet

@Uni_Stuttgart @Cyber_Valley @AlinaRoitberg @LSHVU @hidss4health @AlexanderJaus @JensKleesiek @KITKarlsruhe