Pol Caselles

Deep Learning Research Engineer @Crisalix | PhD Candidate in AI/DL

[Pol Caselles @CasellesPol]

[2/2] We present a method for estimating high-quality 3D face reconstructions from just a few input images. Our hybrid approach combines per-vertex neural optimization with landmark-based guidance, enabling flexible geometry without relying on parametric priors.

[Pol Caselles @CasellesPol]

[1/2]📢 Excited to share that our paper "GLVD: Guided Learned Vertex Descent" has been accepted to #NeurIPS 2025! 🎉

Check them out here:
🔗 Paper: https://t.co/IhseoaeB1g
🔗 Project: https://t.co/NG1NxH6RVj

This is a joint collaboration with Francesc Moreno Noguer.

[marcelo Sanchez @Marcelo5444]

[1/3] I am thrilled to announce that RETHINED: Real-Time High Resolution Image Inpainting On Edge Devices is being presented as an Oral today at #WACV2025!

📅Oral Session 5.4: Generative Models V 14:00

project: https://t.co/59s70en60j

[Pol Caselles @CasellesPol]

(3/3) If you are interested in what we do at @Crisalix check it out here: https://t.co/iNANRfkxpl

This is a joint collaboration with @eram1205 @jaimestetix @GilTriginer @fmorenoguer

#ComputerVision #MachineLearning #artificial_intelligence #AI #CS

[Pol Caselles @CasellesPol]

(1/3) 📢Excited to share that our journal paper "Implicit Shape and Appearance Priors for Few-Shot Full Head Reconstruction" has been accepted to #IEEE Transactions on Pattern Analysis and Machine Intelligence (#TPAMI)

🔗Project: https://t.co/7HuyShejtS

[Pol Caselles @CasellesPol]

(2/3) We present a method to generate high-quality full-head reconstructions from a few input images (down to single view). We also release the H3DS 2.0 dataset to evaluate them in the wild setting.

🔗Dataset: https://t.co/rV7bnuWQVW

[Konstantin Schürholt @k_schuerholt]

🚀 Announcing the Weight Space Learning Workshop at #ICLR2025 @iclr_conf

When neural nets are trained, it’s the weights that embed all the information. Our workshop explores treating these weights as new data modality.

Interested? Join us in Singapore!
https://t.co/4AGREmrbSb

[CrisalixLabs @CrisalixLabs]

👋 Hello @eccvconf ! Crisalix is attending #ECCV2024, if you see @ibrmlk @CasellesPol @Marcelo5444 and want to talk 3D / CV, come say hi.

[Benet Oriol Sàbat @osbenet]

With NeRF-Insert, you can edit a 3D scene with coarse manual masks + text prompt. It also allows different level of control by using a mesh or visual reference for inpainting. Work done at @AmazonScience .

[Pol Caselles @CasellesPol]

We have just released the H3DS 2.0 dataset with up to 59 scenes 🎉🎉. Both the data and the evaluation code are accesible through a Python package.

You can start using it at https://t.co/iA9vXQDYFR

[CrisalixLabs @CrisalixLabs]

Proud to sponsor the @dlbcnai again this year! Looking forward to catching up with the hugely talented community of DL researchers in/from Barcelona. Come by our stand to say hi and meet our team. We’ll also have some cool demos!

[Mariona Carós @MarionaCaros]

Yesterday I presented our work on LiDAR semantic segmentation at the @WiCVworkshop during #CVPR2023. Thanks to everybody who stopped by, and a special thanks to #WiCV for organizing such a nice workshop!

[Pol Caselles @CasellesPol]

Delighted to announce that "SIRA: Relightable Avatars From a Single Image" has been accepted in #WACV2023, @wacv_official !

This is joint work with @eram1205, @jaimestetix, @DocXavi, @fmorenoguer and @GilTriginer.

Project page: https://t.co/MHyeWZBUM7

[Deep Learning Barcelona Symposium @dlbcnai]

#DLBCN 2022 Post-party is over. Thanks to all presenters and attendees for being here, again.

Merry Christmas, bon nadal, and see you again in 2023 !

[Deep Learning Barcelona Symposium @dlbcnai]

Pol Casellas (@CasellesPol) from @Crisalix kicks off the 3D session for full head reconstruction from a single image. #WACV2023 #DLBCN

https://t.co/ILjnNplqGN

[Xavi Giró @DocXavi]

Last day at @la_UPC after 20 years of teaching & research @UPCTelecos @eseiaat_upc. Thanks to everybody who has helped me, especially to my advisor Ferran Marqués and lovely students, who threw me a surprise party��� next to the brand new building of @AmazonScience Barcelona. ❤️

[Jordi Vitrià @bitenmascarado]

PointNet implementation explained visually by Mariona Carós, PhD student at the University of Barcelona. @DataScienceUB https://t.co/nje3ZnsuUZ

[Xavi Giró @DocXavi]

Instructors dinner of our Postgraduate on Artificial Intelligence at Deep Learning at @UPC_School Barcelona. We are already preparing the 11th edition, online, starting on November 2022:

https://t.co/i2ghV23GV9

[UPC School @UPC_School]

📆10/02. New Open Class on Postgraduate in Artificial Intelligence with #DeepLearning.
We will offer an overview of deep learning architectures used for #3D analysis and reconstruction.
🗣️@GilTriginer, Deep Learning Engineer at @Crisalix.
To attend👉https://t.co/OZbbVM39PY

[Benet Oriol Sàbat @osbenet]

SALAI-Net is almost out! Can't wait to publish the results and code. I was able to share an overview of the project at the @dlbcnai during one of the poster sessions.
@DocXavi @_danielmas @alexGioannidis