Alejandro Amat Payá
Portrait of Alejandro Amat Payá

Alejandro Amat Payá

Research Associate, Human Sensing Lab · Robotics Institute, Carnegie Mellon University

I'm a Research Associate at CMU working with Prof. Fernando De la Torre, and an MSCS student at CMU. I hold a dual Bachelor's degree in Electrical Engineering and Computer Science from CFIS-UPC Barcelona.

My work focuses on real-time differentiable rendering, human avatars, 3D reconstruction, and computer vision. I combine research in graphics and machine learning with high-performance engineering.

Differentiable Rendering Human Avatars 3D Reconstruction Gaussian Splatting Computer Vision GPU / HPC
Email GitHub LinkedIn

Experience

  1. Feb 2024 — Now

    Research Associate

    Human Sensing Lab, Carnegie Mellon University

    Real-time neural rendering, human avatars and 3D reconstruction.

  2. Oct 2023 — Apr 2024

    Unity VR Developer

    IESE Business School

    Cross-platform VR app for an IESE / MIT / LBS research collaboration.

  3. May 2023 — Dec 2023

    AI Researcher

    i2CAT Foundation

    Real-time 3D imaging with mmWave radar and intelligent surfaces.

  4. May 2022 — Sep 2022

    Big Data Engineer Intern

    Mango

    Automated data pipelines on AWS for retail analytics.

Publications

Multi-view 3D reconstruction of a basketball game
SIGGRAPH 2025 · ACM TOG

Echoes of the Coliseum: Towards 3D Live Streaming of Sports Events

Junkai Huang*, Saswat Subhajyoti Mallick*, Alejandro Amat, Marc Ruiz Olle, Albert Mosella-Montoro, Bernhard Kerbl, Francisco Vicente Carrasco, Fernando de la Torre

* equal contribution

Paper Project page Video Dataset DOI
mmWave reconfigurable intelligent surface setup
IET Book Chapter · 2025

Location-Based Real-Time Utilization of Reconfigurable Intelligent Surfaces for mmWave Integrated Communication and Sensing

Filip Lemic, Jalal Jalali, Gerard Calvo Bartra, Alejandro Amat, Jakob Struye, Jeroen Famaey, Xavier Costa Perez

IET Advanced Metaverse Wireless Communication Systems

Education

  1. Expected 2028

    M.S. in Computer Science

    Carnegie Mellon University · Pittsburgh, PA

    Coursework: 15-662 Computer Graphics (4.0), 16-720 Computer Vision (4.0), 15-513 Introduction to Computer Systems (4.0), 15-618 Parallel Computer Architecture and Programming (in progress).

  2. Feb 2024 — Jul 2024

    Visiting Research Student · Bachelor's Thesis 10/10

    Robotics Institute, Carnegie Mellon University

    Real-Time Differentiable Rendering with Slang, supervised by Prof. Fernando De la Torre.

  3. 2019 — 2024

    Dual B.Eng. in Electrical Engineering & Computer Science Honors

    CFIS · Universitat Politècnica de Catalunya, Barcelona

    Graduated with honors; ranked 6th of the 2024 graduating class.

Projects

Animated 3D human avatar

Real-Time 3D Human Avatars

Capture and re-animate digital humans from monocular video. Combines 3D Gaussian Splatting, SMPL/FLAME body models, pose estimation, segmentation and canonical optimization with differentiable rendering: <20 min training, 15 FPS live streaming.

3DGSSMPLFLAMEPyTorch
Gaussian splatting scene rendered in Vulkan

Vulkan 3D Gaussian Splatting Renderer

Cross-platform 3DGS engine that breaks CUDA lock-in using Vulkan and GLSL compute: tile-based rasterization, multi-stage radix sort, keyframe animation and Python bindings. 60 FPS on an RTX 3060 Ti, 24–45 FPS on Apple Silicon.

VulkanGLSLC++PyBind
Code
2D Gaussian splatting optimization

Real-Time Differentiable Rendering with Slang

C++ rendering system and shader pipeline built on Slang's automatic differentiation, with custom optimizers and GPU kernels. 116× faster than NVIDIA Falcor for inverse texturing and 4× faster 2D Gaussian Splatting.

SlangC++Autodiff
Code
Path-traced render from Scotty3D

Scotty3D Graphics Pipeline

Software rasterizer (perspective-correct interpolation, mipmapping, supersampling), mesh editor (Loop / Catmull-Clark subdivision, isotropic remeshing) and a BVH-accelerated Monte Carlo path tracer with physically-based materials and HDR environment lighting.

C++Path tracingCMU 15-662
Show
Sobel edge detection comparison

High-Performance CUDA Suite

GPU-accelerated image processing: Sobel edge detection 224,733× faster than NumPy, CUDA-JPEG 120× faster than CPU, and an optimized Multi-Scale SSIM for real-time quality assessment.

CUDApybind11
Code
HP logo

HP SitePrint Verification Library

Led a Python library for the HP SitePrint construction robot: point cloud segmentation, DBSCAN clustering and distance verification with an Intel RealSense D435i. <3% error, published on PyPI.

PythonPoint clouds
PyPI Code Demo
VR museum simulation

Museum VR Simulation

Cross-platform Unity VR experience for researchers at IESE, MIT and LBS: gizmo controls, gyroscope/accelerometer sensor fusion, crowd simulation, spatial audio and Firebase sync. Runs on Google Cardboard.

UnityC#Firebase
Play (web/PC)
mmWave radar heatmap

mmWave Radar & IRS Simulator

MATLAB simulator for mmWave radar heatmap generation and Intelligent Reflecting Surface optimization, enabling multi-user tracking and signal homogeneity analysis for immersive VR.

MATLABRadar
Code
ZeOS kernel output

ZeOS: x86 Kernel

Extended a teaching OS kernel in C and x86 assembly: system call interface, keyboard and clock interrupt handlers, and process management with scheduling and context switching.

Cx86 asm
Code