Alexander Valverde

computer graphics / ucsc

[Alexander Valverde @Alexvalgui]

@allywooww De la necesidad de una ruta sobre investigaciones académicas en teoría y aplicaciones de IA dentro de nuestro sistema de educación superior. Así como la adopción de las mismas en las distintas industrias

[Alexander Valverde @Alexvalgui]

@allywooww Lo peor es que salen ganando todos menos ellos (CAC) y obvio perjudica a LN

[Alexander Valverde @Alexvalgui]

@ASadSisyphus El detalle de los cambios en el empleo por IA tiene que ir orientada hacia la modificación y mejora de las mismas herramientas según las necesidades de cada gremio y con eso poder generar conocimiento local que sea exportable. Tenemos que crear nuestra propia industria

[Alexander Valverde @Alexvalgui]

@ICCVConference @CV4E_ICCV presentation ready!

Great research from different areas.

Pd: A little bit different poster this time, hope to continue working on this interactive projects!

[Alexander Valverde @Alexvalgui]

I benchmarked the method against long-standing Statistical Outlier Removal and Radius Outlier Removal techniques, and evaluated it on three scene reconstruction datasets.

The results show that the proposed filter produces cleaner, more compact point clouds without compromising surface normals or overall scene structure.

The PSNR and SSIM metrics obtained using this optimized point cloud showed quite similar values (and better in some scenes) to the original COLMAP point cloud.

To test this removal filter, I trained 12 scenes from Mip-Nerf 360, Tanks and Temples and DTU on the 2DGS paper https://t.co/8490iRBfi8

I want to thanks @yuyinzhou_cs for all her support during this year. All the experience obtained under her supervision deepened my knowledge and desired to continue a career in computer graphics.

Link to thesis: https://t.co/tgcDEWAmGh

[Alexander Valverde @Alexvalgui]

Today my master's thesis has been officially released!
In this thesis I presented an outlier removal method special for 3D point clouds based on a convex optimization problem.

Gaussian Splatting methods are highly sensitive to the original point cloud. Due to this, is necessary to used a clear object that can express the scene with high-detail and low noise. However, many of these representations have undesirable elements created by mathematical inaccuracies during the triangulation process.

Current methods to clean these representations rely on the use of hyper-parameters that aren't easy to fine-tune, specially for different point cloud scales.

In this research, I build a fast approach used to denoised 3D point clouds created by photogrammetry techniques.

[Alexander Valverde @Alexvalgui]

Very happy to announce that our research “Back Home: A Computer Vision Solution to Seashell Identification for Ecological Restoration” has been accepted at ICCV, 2025 (CV4E Workshop).

https://t.co/endJB2UlMf

Cannot wait to present this project next October in Hawaii

[Alexander Valverde @Alexvalgui]

@surbica Voy a llevarle el “problema” a mi supervisora jaja, a ver si puede salir algo interesante, pero gracias por probarlo!

[Alexander Valverde @Alexvalgui]

@surbica Muy interesante la verdad, veo que es bastante susceptible con la detección de objetos a larga distancia. Ese que hiciste primero (el de la imagen blurreada) puede que sea algo de incluir en en estos modelos de segmentación como forma de guía para detectar este tipo de elementos

[Jon Barron @jon_barron]

One reason generative 3D is hard is that the world has barely any 3D training data. Today, most humans on earth will snap a few photos and write paragraphs of text. In contrast, earth has maybe ~50k 3D artists, and each will make maybe ~1k photoreal 3D models over their career.

[youming.deng @denghilbert]

How can we use wide-FOV cameras for reconstruction?

We propose self-calibration Gaussian Splatting that jointly optimizes camera parameters, lens distortion, and 3D Gaussian representations to directly reconstruct from a set of wide-angle captures.

page: https://t.co/OQ4C20VsvU

[MrNeRF @janusch_patas]

About 30 related papers today!!! Let the 3DGS floodgates burst open!

[Cerveza Imperial @DespegaImperial]

Cada año, millones de conchas son arrancadas de nuestras costas. Esta es la historia de cómo las devolvimos al lugar donde pertenecen. 🌊 🐚 ¡Compartí este video y ayudanos a seguir protegiendo nuestras playas! 📽️✨ Conocé más en https://t.co/5G1ppuKGxY

#DeVueltaACasa

[Alexander Valverde @Alexvalgui]

Have you ever ask yourself how can Images can be analyzed to obtain key information?
Look my recent medium article about how to extract important information from images using wavelets and the fourier transform!!
https://t.co/XFPDSfA7LT
#computervision #python #deeplearning

[Alexander Valverde @Alexvalgui]

Se podría penalizar el mal uso de una herramienta pero eso no sería propiamente al modelo sino a las personas que utilizan estas herramientas para generar algún daño.
Si creen que pueden regular los modelos como tal, mejor vayan a leerse los papers para evitar estos discursos.

[Barra de Prensa @barradeprensa]

Bloque de diputados de distintas fracciones anuncian la presentación de un proyecto de ley hecho con ChatGPT para regular la Inteligencia Artificial.

[Alexander Valverde @Alexvalgui]

@gabrielpeyre Does it mean that Wavelet transform can be used to generate a “common image” to obtain a general idea about some relations between a bunch of pics that maybe have Non Linear relationships?

Inferring that it will select the best coefficients of the decompositions or no?

[Alexander Valverde @Alexvalgui]

See my recent article of SVD technique to decompose a faces matrix to reconstruct faces based on the more important components of the matrix.
Based on the Data-Driven Science and Engineering book from @eigensteve and Nathan Kutz

https://t.co/QvXs3LBWeD
@svd @pca @ml @faces

[Alexander Valverde @Alexvalgui]

Best explanation ever of why gradient descent exists!

[Tivadar Danka @TivadarDanka]

In machine learning, we take gradient descent for granted. We rarely question why it works.

What's usually told is the mountain-climbing analogue: to find the valley, step towards the steepest descent.

But why does this work so well? Read on.

[Alexander Valverde @Alexvalgui]

I want to remark in the computational cost of the 175B SFT respect other models, cost is a very important topic that needs to be consider in the future to develop better models

[Alexander Valverde @Alexvalgui]

Take a look at this paper if you want to know more about LLM, its a very good work developed by OpenAI RS about many of their models and the way were created, a nice document to take a look and understand more about them https://t.co/S4dxYtYnjX #OpenAI #InstructGPT #ChatGPT