I applied for a visa 2.5 months ago, and I still haven't received it. It's frustrating that the MICCAI registration won't refund at least half of my registration fees, which total over $800! 🙁 #VisaDelay #MICCAIRegistration #MICCAI2023

Come chat at the poster session 4pm Monday! Joint work with @Warvito, @paulwright74, @PTudosiu, Parashkev Nachev, Seb Ourselin, @mjorgecardoso among others! Paper: https://t.co/AX2bMHSRGF Code:

Will be presenting our work on diffusion models for fully 3D out-of-distribution at #MICCAI2023. We reconstruct inputs from various t-values, and use these recon errors across the t-chain to flag OOD images. We show this works well and can be extended to 3D data using LDMs.

🚀 Join us at on Sept 26th, 3:30-4:30pm at Union Mills, Leeds, or online. Don't miss #LDF2023, featuring Joe Batt and Alex Woodhead as they join @marksgraham_, and @paulwright74 from @KingsCollegeLon to explore FLIP's inner workings!🧠 🔗 Event:

And check out the codebase!

Are you interested in image-to-image applications with medical images? Check out our latest article on @TDataScience, where we walk you through the entire process of training ControlNets to control Latent Diffusion Models. https://t.co/KmOCGhdpHg #MedicalAI #GenerativeAI #MONAI

Lastly thanks to the VAND organizing team - @ThomasBrox, @TobyBreckon, Philipp Seeböck, Paul Bergmann, Latha Pemula

And the performance improvement becomes more pronounced when we consider larger images - 64^2 and 128^2 medical images in this case.

This approach beats both reconstruction and likelihood-based approaches to OOD detection on simple 32x32 images from computer vision benchmarks.

We can noise an input to several different levels, reconstruct each with a DDPM, and then take the mean recon error across all these different bottlenecks.

Reconstruction-based approaches to OOD don’t work. If your model’s bottleneck is too small, it can't reconstruct ID data, too big, even OOD data is well-reconstructed. With DDPMs, the bottleneck is not part of the model architecture, but controlled by the level of noise applied.

Our work applying diffusion models to out-of-distribution detection has been accepted at the CVPR workshop on Visual Anomaly Detection. Paper: https://t.co/y9kKqsLOlJ Code: https://t.co/32dpwUeIhz with @mjorgecardoso, @Warvito, @PTudosiu, Seb Ourselin, Parashkev Nachev

2x Senior Research Fellow posts: We have recently released MONAI Label (400+ Github ⭐️) and Generative Models (200+ Github ⭐️), part of @ProjectMONAI. Now we’re growing. Want to join? MONAI Label: https://t.co/G8Py1Y2Z9X MONAI Generative: https://t.co/EZfCqgT6aC Join us!

We are excited to announce that SASHIMI 2023 will run jointly with the #SynthRAD2023 challenge! Check the workshop ( https://t.co/b3oabhQ7g0) or the challenge ( https://t.co/K7CX6d3lem) websites for more info!

The ControlNet tutorial in MONAI Generative walks you controlling the shape of brain MRI samples with a supplied brain mask. Check it out!

Hi all! Our work with MONAI Generative Models was featured as the AI research of the month on Computer Vision News @RSIPvision by @maricaS8 ! https://t.co/BF8xNIhi1P

PhD position: ‘Scaling up human-AI interactions: application to medical data labelling’ Re-advertising an industry funded PhD project with Siemens Healthineers and @ProjectMONAI. Get in touch. Eligibility: Home/UK students Closing date: 17th Apr https://t.co/UBXkloSR5d

Me, @Warvito and @vir_fgcarmena will be talking about MONAI Generative ( https://t.co/elRLlxDnEc) at NVIDIA GTC, starts in five mins! https://t.co/0sL7o4FmXj

It is finally out! Check out our project to bring generative models to @ProjectMONAI and help develop several medical imaging tasks such as image synthesis, anomaly detection, denoising, super-resolution, MRI reconstruction, and more!

Want to do a industry-funded PhD on active learning and human-AI interaction, aiming to improve medical imaging data labelling? Apply below (international applicants, deadline 1st March) Joint with @SiemensHealth & building on @ProjectMONAI Label https://t.co/UBXkloSR5d