iiinova — where your digital self keeps evolving

Reposted by Fernando Pérez-García

Moira Donegan @moiradonegan.bsky.social · 23/02/2026

Why do I have to pretend that I'm going to print something in order to save it as a PDF. Why do I have to engage in a little ruse.

334191442875

Fernando Pérez-García @fepegar.com · 30/01/2026

...Maria Teodora Wetscherek, Klaus Maier-Hein, Panagiotis Korfiatis, @valesalvatelli.bsky.social, Javier Alvarez-Valle. 🧵12/12

000

Fernando Pérez-García @fepegar.com · 30/01/2026

Extra kudos to Tassilo Wald, who led the execution! Thanks to everyone else involved in the project: @ibrahimethem.bsky.social, Yuan (William) Gao, @sambondtaylor.bsky.social, Harshita Sharma, @maxilse.bsky.social, Cynthia Lo, Olesya Melnichenko, @anton-sc.bsky.social, Noel Codella... 🧵11/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

COLIPRI is very easy to install and use! Just run `pip install colipri` and paste this snippet (from aka.ms/colipri) to get started. I'm looking forward to seeing what the community will build on top of our model. Get in touch if you have questions or feedback! 🧵10/12

aka.ms

microsoft/colipri · Hugging Face

We’re on a journey to advance and democratize artificial intelligence through open source and open science.

100

Fernando Pérez-García @fepegar.com · 30/01/2026

We used nnSSL (Wald et al., ICCV 2025) for training, TorchIO (@fepegar.com et al., CMPB 2021) for preprocessing and augmentation, nnU-Net (Isensee et al., Nature Methods 2021) for segmentation, and nifti-zarr-py for efficient patch loading from during cloud training. 🧵9/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

COLIPRI is generally superior to concurrent methods across all tasks. This is particularly clear when plugging an MLLM on top of our vision backbone. Our models are particularly stronger at clinical metrics, which are most relevant in practice. 🧵8/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

To overcome this domain shift, we introduce an Opposite Sentence Loss (OSL), a simple but effective mechanism that complementes the contrastive loss and improves our metrics substantially. 🧵7/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

Medical reports are often very long and most sentences describe what is *not* in the scan. However, for zero-shot classification, users tend to use very short prompts, such as "Lung nodules" and "No lung nodules". 🧵6/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

We generate reports during training to ensure that the vision encoder extracts from the image all the information that would be needed for reporting, similar to CapPa (@mtschannen.bsky.social et al., NeurIPS 2023). 🧵5/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

We resampled the volumes to 2-mm isotropic spacing using and used an input size of 160^3. We randomly shuffled and shortened sentences in the reports used for contrastive alignment. We initialised our encoder from CXR-BERT (Boecking, @naotous.bsky.social et al., ECCV 2022). 🧵4/12

110

Fernando Pérez-García @fepegar.com · 30/01/2026

We first pre-train our encoder only on images (no reports) sourced from different datasets, using a 3D MAE (Wald et al., CVPR 2025). This allows us to leverage more training data, as we did for Rᴀᴅ-DINO (@fepegar.com et al., Nature Machine Intelligence 2025). 🧵2/12

100

Fernando Pérez-García @fepegar.com · 30/01/2026

There is not a lot of paired 3D medical image–report data out there. A pioneering example is CT-RATE (@iethemhamamci, arxiv.org/abs/2403.17834), with samples from 21k patients, a number much smaller than what we see in the natural imaging domain. 🧵1/12

arxiv.org

Developing Generalist Foundation Models from a Multimodal Dataset for 3D Computed Tomography

Advancements in medical imaging AI, particularly in 3D imaging, have been limited due to the scarcity of comprehensive datasets. We introduce CT-RATE, a public dataset that pairs 3D medical images wit...

100

Fernando Pérez-García @fepegar.com · 30/01/2026

We are excited to release the weights of @msftresearch.bsky.social's COLIPRI, our 3D vision–language encoder for chest CT scans, on @hf.co 🤗 Model: aka.ms/colipri Demo: aka.ms/colipri-demo Paper: aka.ms/colipri-paper Why does COLIPRI matter? 🧵0/12 👇

132

Reposted by Fernando Pérez-García

Andrew Heiss @andrew.heiss.phd · 15/12/2025

Grading and googling hallucinated citations, as one does nowadays, and now that LLMs have been around for a while, I've discovered new horrors: hallucinated journals are now appearing in Google Scholar with dozens of citations bc so many people are citing these fake things

13039551264

Reposted by Fernando Pérez-García

Gautam Kamath @gautamkamath.com · 07/12/2025

This article frames the problem as [AI?] slop. Which is a problem, but not the main one here. This is an issue with authorship norms and practices. A single individual putting their name on hundreds of (workshop) papers they admitted they had little part in.

3213

Reposted by Fernando Pérez-García

Daniel Coelho de Castro @dccastr0.bsky.social · 07/12/2025

Excited to speak shortly at the Medical Imaging at @euripsconf.bsky.social workshop in Copenhagen, where I'll share some insights from our @msftresearch.bsky.social team's journey "From medical image interpretation to scientific discovery" over the past couple of years.

Opening slide for a presentation titled "From medical image interpretation to scientific discovery"

022

Reposted by Fernando Pérez-García

Lukas C. H. @gewoonlukas.bsky.social · 26/11/2025

SpaceX is ready for its next Transporter mission! With 140 satellites onboard, this is the largest Transporter mission since Transporter-1 in 2021, which carried 143 satellites. Here's my identification attempt. Launch is scheduled for NET 18:19 UTC.

269737

Reposted by Fernando Pérez-García

Python Software Foundation @python.org · 27/10/2025

TLDR; The PSF has made the decision to put our community and our shared diversity, equity, and inclusion values ahead of seeking $1.5M in new revenue. Please read and share. pyfound.blogspot.com/2025/10/NSF-... 🧵

python.org

The official home of the Python Programming Language

12363772727

Reposted by Fernando Pérez-García

Max Ilse @maxilse.bsky.social · 23/09/2025

🩻Excited to share our latest preprint: “Data Scaling Laws for Radiology Foundation Models” Foundation vision encoders like CLIP and DINOv2 have transformed general computer vision, but what happens when we scale them for medical imaging? 📄 Read the full preprint here: arxiv.org/abs/2509.12818

arxiv.org

Data Scaling Laws for Radiology Foundation Models

Foundation vision encoders such as CLIP and DINOv2, trained on web-scale data, exhibit strong transfer performance across tasks and datasets. However, medical imaging foundation models remain constrai...

152

Reposted by Fernando Pérez-García

Amnesty International UK @amnestyuk.bsky.social · 22/09/2025

Recognition is no doubt significant but it will be a hollow gesture if the UK does not also seek to end Israel's genocide, illegal occupation, and system of apartheid against the Palestinian people. 🧵1/3

Photo with Keir Starmer and copy that reads: RECOGNITION OF A PALESTINIAN STATE IS A HOLLOW GESTURE WITHOUT MEANINGFUL ACTION TO END ISRAEL'S GENOCIDE, APARTHEID & OCCUPATION

511560

Reposted by Fernando Pérez-García

Christopher Owen @oysta.au · 21/09/2025

Yeah man we should really fight back by staying on X

256169914055

Fernando Pérez-García @fepegar.com · 21/08/2025

Having some fun with DINOv3 and PCA! Although I'm not happy my nose has such a low foreground probability :D

Grid of PCA visualizations of my face, computed with DINOv3.

Left: a photo of myself. Right: a heatmap representing the foreground probability of each region.

060

Reposted by Fernando Pérez-García

Sam Rose @samwho.dev · 17/08/2025

Email addresses are very simple, and you will score highly in this quiz. e-mail.wtf

e-mail.wtf

Email is Easy

Everyone knows what an email address is, right?

38274128

Reposted by Fernando Pérez-García

Max Seitzer @maxseitzer.bsky.social · 14/08/2025

Introducing DINOv3 🦕🦕🦕 A SotA-enabling vision foundation model, trained with pure self-supervised learning (SSL) at scale. High quality dense features, combining unprecedented semantic and geometric scene understanding. Three reasons why this matters👇

2269

Reposted by Fernando Pérez-García

Merriam-Webster @merriam-webster.com · 08/08/2025

We’re not sure who needs to hear this, but ‘blueberry’ has two b’s.

2337128918

Reposted by Fernando Pérez-García

Olivia Guest · Ολίβια Γκεστ @olivia.science · 29/07/2025

Boiling here at home in Cyprus but I put the finishing touches a couple of days ago on this preprint: What Does 'Human-Centred AI' Mean? doi.org/10.48550/arX... Wherein I analyse HCAI & demonstrate through 3 triplets my new tripartite definition of AI (Table 1) that properly centres the human. 1/n

title and abstract from https://arxiv.org/pdf/2507.19960

table 1 from https://arxiv.org/pdf/2507.19960

11419122

Reposted by Fernando Pérez-García

Nature @nature.com · 23/07/2025

Scientists overwhelmingly recognize the value of sharing null results, but rarely publish them in the research literature go.nature.com/450KElr

go.nature.com

Researchers value null results, but struggle to publish them

Survey finds that fear of reputational harm and a lack of support and publication platforms are among respondents’ key concerns.

513766

Reposted by Fernando Pérez-García

Ellie Huxtable @ellie.wtf · 23/07/2025

i love it when people make PRs really easy to review ❤️ feels really bad when there's a PR that's been open for a while, but i know i can't easily do a good job of reviewing it, so it's constantly "later"

2221

Reposted by Fernando Pérez-García

DrBreaky @drbreaky.bsky.social · 21/07/2025

New 2 year position as an Imaging Data Infrastructure Specialist Located a the Hunter Medical Research Institute (HMRI) & the University of Newcastle (UoN), in partnership with the National Imaging Facility (NIF), Apply here: www.seek.com.au/job/85537666

0178

Reposted by Fernando Pérez-García

Kosta Derpanis @csprofkgd.bsky.social · 19/07/2025

A #computervision researcher at #ICML2025

3657

Reposted by Fernando Pérez-García

Stephanie Hyland @hylandsl.bsky.social · 18/07/2025

New work from my team! arxiv.org/abs/2507.12950 Intersecting mechanistic interpretability and health AI 😎 We trained and interpreted sparse autoencoders on MAIRA-2, our radiology MLLM. We found a range of human-interpretable radiology reporting concepts, but also many uninterpretable SAE features.

arxiv.org

Insights into a radiology-specialised multimodal large language model with sparse autoencoders

Interpretability can improve the safety, transparency and trust of AI models, which is especially important in healthcare applications where decisions often carry significant consequences. Mechanistic...

1114

Reposted by Fernando Pérez-García

Journal of Open Source Software @joss-openjournals.bsky.social · 11/07/2025

Just published in JOSS: 'Lighter: Configuration-Driven Deep Learning' doi.org/10.21105/joss.08101

021

Reposted by Fernando Pérez-García

Microsoft Research @msftresearch.bsky.social · 26/06/2025

The world’s first multimodal, bilingual radiology dataset could reshape the way radiologists and AI systems make sense of X-rays. PadChest-GR, developed by the University of Alicante with Microsoft Research, has the potential to advance research across the field for years to come. msft.it/6013SLDYZ

073

Reposted by Fernando Pérez-García

Daniel Coelho de Castro @dccastr0.bsky.social · 26/06/2025

📣 Our new PadChest-GR benchmark in collaboration with @ua.es and MedBravo is published in @ai.nejm.org: ai.nejm.org/doi/full/10....! 📰 Also check out our blog post for why we're so excited about it: www.microsoft.com/en-us/resear... 💾 The dataset can be downloaded here: bimcv.cipf.es/bimcv-projec...

ai.nejm.org

PadChest-GR: A Bilingual Chest X-Ray Dataset for Grounded Radiology Report Generation

Artificial intelligence (AI)–powered radiology report generation (RRG) aims to create free-text radiology reports from clinical imaging. Grounded radiology report generation (GRRG) augments RRG by ...

062

Reposted by Fernando Pérez-García

NEJM AI @ai.nejm.org · 03/07/2025

A novel dataset designed to train and evaluate grounded report generation models from chest x-ray images includes comprehensive sentence-level bounding-box annotations for all clinically relevant findings in each image. Learn more: nejm.ai/44rk1a1 @dccastr0.bsky.social #AI #MedSky #MLSky

Figure 1. Example of a Grounded Report from PadChest-GR.

044

Reposted by Fernando Pérez-García

pr0grammerhum0r.bsky.social @pr0grammerhum0r.bsky.social · 27/06/2025

codingAssistantsAreJustCasinosForProgrammers

081

Fernando Pérez-García @fepegar.com · 27/06/2025

¿No se puede multar también?

000

Fernando Pérez-García @fepegar.com · 26/06/2025

Congratulations/enhorabuena to @dccastr0.bsky.social and the rest of the team 👏 One important step towards usability of AI in real-world radiology applications.

060

Reposted by Fernando Pérez-García

conda-forge @conda-forge.org · 12/06/2025

Demystifying conda packages!

021

Reposted by Fernando Pérez-García

Kosta Derpanis @csprofkgd.bsky.social · 18/05/2025

A new use of the asterisk in the paper author list for credit assignment

841266

Reposted by Fernando Pérez-García

François Fleuret @francois.fleuret.org · 28/04/2025

I asked "on the other platform" what were the most important improvements to the original 2017 transformer. That was quite popular and here is a synthesis of the responses:

420643

Reposted by Fernando Pérez-García

Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 26/04/2025

Some of the anti-AI stuff feels a bit like when people would say "don't use Wikipedia as a source." It's just like anything else, a piece of information that you weigh against multiple sources and your own understanding of its likely failure modes

55444838

Reposted by Fernando Pérez-García

Sander Dieleman @sedielem.bsky.social · 25/04/2025

One weird trick for better diffusion models: concatenate some DINOv2 features to your latent channels! Combining latents with PCA components extracted from DINOv2 features yields faster training and better samples. Also enables a new guidance strategy. Simple and effective!

0284

Reposted by Fernando Pérez-García

Mauro Entrialgo @tyrexito.bsky.social · 21/04/2025

Al papa lo que es del papa.

462748887

Fernando Pérez-García @fepegar.com · 21/04/2025

You’ll need to create an account to view your quiz score.

000

Reposted by Fernando Pérez-García

Félix López Luis 🇺🇦 @flopezluis.bsky.social · 16/04/2025

I wrote this to help my team build an Operational Excellence mindset. I actually shared it internally with the whole company. Sharing it here in case it's useful to someone: docs.google.com/document/d/1...

docs.google.com

Operational Excellence

Operational Excellence Operational excellence is a mindset and set of practices focused on delivering highly reliable systems with continuous improvement. It blends engineering quality, observability,...

42612