Joaquin Gajardo

Doctoral Researcher at ETH Zürich

I am a Doctoral Researcher at ETH Zürich working on spatial and generative AI applications for agriculture. My research focuses on semantic 3D and 4D reconstruction of plant growth to support precision agriculture and crop breeding in the context of climate change. Previously, I worked as a Data Scientist at Empa on AI for Earth Observation, numerical simulation and mobile app development. I received my MSc from EPFL and my BSc and Engineer title from PUC Chile (top-ranked in Latin America).

3D reconstruction Gaussian Splatting Neural Fields Scene Understanding Generative Models Computer Vision Deep learning Robot Learning Python PyTorch Docker

Education

Ph.D. candidate in Computer and Agricultural Sciences

ETH Zürich, Switzerland

2023 - Present

Topic: Organ-aware 4D reconstruction and modelling of plants.

M.Sc. in Environmental Sciences and Engineering

EPFL, Switzerland

2018 - 2021

⭐ Thesis on Neural Controlled Differential Equations awarded the CSD Ingénieurs Prize.

B.Sc. and Professional Engineer in Industrial Engineering

Pontificia Universidad Católica de Chile, Chile

2012 - 2017

⭐ Distinguished result in degree exams: Specialization (Top 1%) and Fundamentals of Engineering (Top 10%).

Experience

Data Scientist

Empa – Swiss Federal Laboratories for Materials Science and Technology, Switzerland

2021 - 2023

Technical lead of the Your Virtual Cold Chain Assistant project (helped secure initial funding for $1M), collaborating with BASE on the design, development and testing of a mobile app for cold chain optimization in developing countries. Conducted applied research on large-scale agriculture mapping using satellite imagery and deep learning. Supervised data science interns and managed GPU infrastructure.

Intern

Empa – Swiss Federal Laboratories for Materials Science and Technology, Switzerland

2020

Developed a 3D digital twin software in Python for real-time food quality prediction in cold storage, implementing custom numerical PDE/ODE solvers for physics-based thermal transport modeling. The software was built for and deployed by an industrial partner.

Publications

Wheat3DGS: In-field 3D Reconstruction, Instance Segmentation and Phenotyping of Wheat Heads with Gaussian Splatting

Daiwei Zhang*, Joaquin Gajardo*, Tomislav Medic, Isinsu Katircioglu, Mike Boss, Norbert Kirchgessner, Achim Walter, Lukas Roth

Proceedings of the Computer Vision and Pattern Recognition Conference (CVPR) Workshops 2025

* Shared first authorship

First-of-its-kind method for automatic 3D instance segmentation of wheat heads in field conditions, enabling precise measurement of key morphological traits such as length, width and volume.

project page paper arXiv code

Evaluating the role of training data origin for country-scale cropland mapping in data-scarce regions: A case study of Nigeria

Joaquin Gajardo, Michele Volpi, Daniel Onwude, Thijs Defraeye

ISPRS Open Journal of Photogrammetry and Remote Sensing 2025

We investigate the role of training data quantity and origin for accurate country-scale cropland mapping at 10m resolution.

paper arXiv code demo

Projects

Towel Folding with VLAs and Flow Matching

ETH Zürich

Apr 2026 – May 2026

Project for Robot Learning course at ETH. Robot towel folding with fine-tuned flow-matching based VLA policies (SmolVLA and MolmoAct2) on SO101 robot arms. Open-sourced models, datasets, and training pipeline.

3D Timelapses of Plant Dynamics with 3D Gaussian Splatting

ETH Zürich

Mar 2024 – Jun 2024

Project for 3D Vision course at ETH. Created 3D timelapses of plant dynamics (day-night movement and growth) with aligned per-timestep 3D scene reconstructions from multi-view images with 3D Gaussian Splatting. Supervised by Dr. Sergey Prokudin.

Coldtivate

Empa

2021-2023

Led the design and development of an open-source AI- and simulation-powered mobile application at Empa to optimize cold chain logistics for perishable goods, enhancing efficiency and reducing waste in supply chains.

project page video article poster code

MSc thesis

Exchange semester at ETH Zürich, PRS lab (Ecovision)

2020-2021

Neural Controlled Differential Equations for crop classification on time series of satellite images. Proposed a new stacked architecture and earned CSD Ingénieurs prize. Supervised by Prof. Jan Wegner (ETH/UZH) and Prof. Devis Tuia (EPFL).

thesis code

Supervised Students

Patrick Eugster — BSc Thesis, ETH Zürich, 2026

Juliane Mercoli & Kevin Zihlman — Data Science Lab, ETH Zürich, 2026

Claudio Abart — MSc Thesis, ETH Zürich, 2025

Daiwei Zhang — MSc Semester Project, ETH Zürich, 2024

Sélène Ledain — Internship (EPFL MSc student), Empa, 2022

Danya Li — Internship (EPFL MSc student), Empa, 2021