Fabian Perez

Just me. Computer science student at UIS, i like to train neural nets
ig:fabian.przz

[Fabian Perez @fabianprzz]

@jxmnop Ewwww 🤮

[Fabian Perez @fabianprzz]

It was a pleasure to present at @ICCVConference . Great ideas and insightful discussions during the poster session today, thanks to everyone who stopped by!

[KAUST Computer Vision Lab (IVUL) @KaustVision]

🌟Excited to share that our IVUL Lab @KaustVision has seven papers accepted at @ICCVConference 2025!
We'll be presenting them this month in Hawaii 🌺🌴

👀 Check out the details in the images below.
Hope to see you there! Big thanks to all our amazing collaborators 🤝
#ICCV2025

[SoccerNet @soccernet_org]

🚨 The SoccerNet Challenge 2025 Report is out! 🚨
👉 https://t.co/Dkj1P53ibf
Thanks to all teams pushing the limits of AI in football ⚽📊

Stay tuned: SoccerNet Challenges 2026 will be unveiled progressively in the next 2 months!

Join us on Discord 👉 https://t.co/OWpRf2UThn

[Fabian Perez @fabianprzz]

@https_apricot ah pero la de mejorar la red de movistar si no se la saben

[Fabian Perez @fabianprzz]

📄 Paper: https://t.co/sSBb4KnPDy
🌐 Project page: https://t.co/8KqCNoZI6u
💻 Code: https://t.co/3h7p3T2CwY
📊 Dataset: https://t.co/Rz3jX5ZmgP

See you in Hawaii! 🏝️ @ICCVConference

[Fabian Perez @fabianprzz]

🚀 Excited to share that our paper “UnMix-NeRF: Spectral Unmixing Meets Neural Radiance Fields” has been accepted to #ICCV2025 !

[Fabian Perez @fabianprzz]

UnMix-NeRF integrates spectral unmixing with NeRFs to enable simultaneous hyperspectral view synthesis and unsupervised material segmentation. Our approach captures the intrinsic spectral properties of materials, allowing material segmentation and scene editing.

[Nick Jiang @nickhjiang]

Vision transformers have high-norm outliers that hurt performance and distort attention. While prior work removed them by retraining with “register” tokens, we find the mechanism behind outliers and make registers at ✨test-time✨—giving clean features and better performance! 🧵

[Jack Morris @jxmnop]

the transformer architecture might be the most important invention of the 21st century

for years people talked about exploding and vanishing gradients,
optimizers, learning rate schedules, etc

none of that matters much anymore thanks to transformers. computers just learn now

[Fabian Perez @fabianprzz]

Excited to share that I won the @googlecloud Hackathon Vertex AI 🎉. Such an incredible experience and opportunity. 🥹🇨🇴🙏

[Google Cloud Tech @GoogleCloudTech]

Leave @fabianprzz a 👏🏾 in the comments for his win at the Vertex AI Agent Builder Hackathon in the Productivity Booster category!

Then, learn about his Yipao Agent and how he build it on Google Cloud → https://t.co/VTrdbU7RQP

[Fabian Perez @fabianprzz]

@btibor91 hi tibor, why don't you use a public scraper to keep track of the different versions of chatgpt?
just like https://t.co/RzljtfxiBi

[Fabian Perez @fabianprzz]

@sama 🍓

[Sam Altman @sama]

When you learn that you can learn anything, you are willing to take on anything.

And when you learn that you can win with execution speed alone, you are willing to take on anyone.

[Fabian Perez @fabianprzz]

@tszzl It always has been

[Rishabh @rishabh16_]

3-WL spectral GNN with laplacian encodings, virtual nodes, and graph rewiring // 2 transformer layers

[Fabian Perez @fabianprzz]

@labelmeai Which model is being used?

[Fabian Perez @fabianprzz]

🔗Dive into our research:

Paper: https://t.co/KHK2QdYhd4

Website: https://t.co/YHcFEgbrjU

Code is available at: https://t.co/EGQE7ChvCl #cvpr #materialsegmentation #ComputerVision

#cvpr #materialsegmentation #computervision @CVPR

[Fabian Perez @fabianprzz]

🚀Excited to announce that our paper "Beyond Appearances: Material Segmentation with Embedded Spectral Information from RGB-D imagery" has been presented at @_LXAI Research workshop at #CVPR2024 ! Collaborating with Hoover Rueda-Chacón on this was incredible

[Fabian Perez @fabianprzz]

🌟This work introduces a framework that embeds spectral information within encoder-decoder architectures for material segmentation using RGB-D images. We've also tested this approach on consumer devices like the iPad Pro using its LiDAR sensor, capturing real-world scenes