Juan Castorena

Dr. Juan Castorena


Research Focus & Metrics

Academic Career Length

7Years (2020-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
6 January 2025 in Science of Remote Sensing
Verified
Journal Article
Open access
JCJuan Castorena
Juan Castorena
LDL. Turin Dickman
L. Turin Dickman
AKAdam J. Killebrew
Adam J. Killebrew
JGJames R. Gattiker
James R. Gattiker
RLRod Linn
Rod Linn
ELE. Louise Loudermilk
E. Louise Loudermilk

11 April 2025 in Proceedings of the AAAI Conference on Artificial Intelligence
Verified
Journal Article
Open access
PDF file
GZGeigh Zollicoffer
Geigh Zollicoffer
MVMinh N. Vu
Minh N. Vu
BNBen Nebgen
Ben Nebgen
JCJuan Castorena
Juan Castorena
BABoian S. Alexandrov
Boian S. Alexandrov
MBManish Bhattarai
Manish Bhattarai
This work presents an information-theoretic examination of diffusion-based purification methods, the state-of-the-art adversarial defenses that utilize diffusion models to remove malicious perturbations in adversarial examples. By theoretically characterizing the inherent purification errors associated with the Markov-based diffusion purifications, we introduce LoRID, a novel Low-Rank Iterative Diffusion purification method designed to remove adversarial perturbation with low intrinsic purification errors. LoRID centers around a multi-stage purification process that leverages multiple rounds of diffusion-denoising loops at the early time-steps of the diffusion models, and the integration of Tucker decomposition, an extension of matrix factorization, to remove adversarial noise at high-noise regimes. Consequently, LoRID increases the effective diffusion time-steps and overcomes strong adversarial attacks, achieving superior robustness performance in CIFAR-10/100, CelebA-HQ, and ImageNet datasets under both white-box and grey-box settings.

18 March 2026 in ISPRS Journal of Photogrammetry and Remote Sensing
Verified
Journal Article
Closed access
JCJuan Castorena
Juan Castorena
ELE. Louise Loudermilk
E. Louise Loudermilk
Scott PokswinskiScott Pokswinski
Scott Pokswinski
RLRodman Linn
Rodman Linn


27 April 2026 in Preprints.org
Verified
Preprint
Open access
PDF file
AMAnthony Marcozzi
Anthony Marcozzi
JTJohnathan Tenny
Johnathan Tenny
DMDaithi Martin
Daithi Martin
JCJuan Castorena
Juan Castorena
ZCZachary Crennen
Zachary Crennen
SHSamuel Hillman
Samuel Hillman
Accurate three-dimensional representations of tree structure are essential for fire modeling, radiative transfer simulation, synthetic data generation, and digital twins of forests, yet detailed 3D structure is rarely available at required scales. Current approaches approximate crowns with smooth geometric primitives, discarding the clumping, gaps, and irregular branching present in real trees. We present TreeFlow, a conditional flow matching model that generates realistic 3D tree point clouds from species, acquisition platform, and height. The model uses a transformer trained on real laser scanning data from the FOR-species20K benchmark to learn a velocity field transporting samples from a Gaussian distribution to the source data distribution. We evaluate generation quality by comparing conditioning and distributional fidelity metrics to scans of real trees. Generated trees match or approach the intra-class baseline on five of six metrics, with a Chamfer distance of 0.581 m versus 0.559 m for real trees of the same genus and height class. Performance is strongest below 25 m and degrades with increasing height. TreeFlow is the first flow matching model to produce 3D tree point clouds from scalar inventory attributes using real laser scanning data.

5 July 2026 in Remote Sensing
Verified
Journal Article
Open access
PDF file
AMAnthony Marcozzi
Anthony Marcozzi
JTJohnathan Tenny
Johnathan Tenny
DMDaithi Martin
Daithi Martin
JCJuan Castorena
Juan Castorena
ZCZachary Crennen
Zachary Crennen
SHSamuel Hillman
Samuel Hillman
Accurate three-dimensional representations of tree structure are essential for fire modeling, radiative transfer simulation, synthetic data generation, and digital twins of forests, yet detailed 3D structure is rarely available at required scales. Current approaches approximate crowns with smooth geometric primitives, discarding the clumping, gaps, and irregular branching present in real trees. We present TreeFlow, a conditional flow matching model that generates realistic 3D tree point clouds from species, acquisition platform, and height. The model uses a transformer trained on real laser scanning data from the FOR-species20K benchmark to learn a velocity field transporting samples from a Gaussian distribution to the source data distribution. We evaluate generation quality by comparing conditioning and distributional fidelity metrics to scans of real trees. Generated trees match or approach the intra-class baseline on five of six metrics, with a Chamfer distance of 0.581 m versus 0.559 m for real trees of the same genus and height class. Performance is strongest below 25 m and degrades with increasing height. TreeFlow generates individual-tree point clouds conditioned on scalar inventory attributes using a model trained entirely on real laser scanning data.