I am a PhD student in Electrical & Computer Engineering at Johns Hopkins University. I work with Professor Rama Chellappa and Dr Vishal Patel in AIEM and VIU Labs.

My research lies at the intersection of multimodal foundation models, large vision–language models, generative models, and geometric computer vision. I develop correspondence-aware architectures that combine semantic priors from LLMs with flow-based and implicit representations, enabling precise registration, structural reasoning, and robust alignment and temporal modeling in remote sensing and world-modeling applications.

Since 2024, I have also been part of the WRIVA program at JHU, working on wide-area visual recognition, localization, and 3D reconstruction.

Outside of research, I enjoy film, yoga, creative writing, and learning new languages.

Email / LinkedIn / Google Scholar / GitHub / CV

News

• Nov 2025 — Two first-author papers accepted to WACV 2026.
• 2024 — Joined the AIEM Lab and started working on the WRIVA program at JHU.
• Fall 2024 & 2025 — Serving as a teaching assistant for Medical Imaging Systems (EN.520.432) with Prof. Jerry Prince.

Research

Morphing Through Time: Diffusion-Based Bridging of Temporal Gaps for Robust Alignment in Change Detection

Anita Madani, Vishal M. Patel — WACV 2026 (to appear)

Remote sensing change detection often suffers from severe misalignment when image acquisitions are separated by large seasonal or multi-year gaps. This paper introduces a diffusion-based semantic morphing pipeline that synthesizes intermediate “bridging” frames between bi-temporal images, enabling robust, stepwise correspondence estimation. The generated morphs guide RoMa-based dense registration, followed by a lightweight U-Net that produces a high-fidelity warp preserving true structural changes. Experiments on LEVIR-CD, WHU-CD, and DSIFN-CD demonstrate consistent improvements in both registration accuracy and downstream change detection across multiple backbones, highlighting the generality and effectiveness of diffusion-assisted temporal alignment.

DiffRegCD: Diffusion-Based Registration for Remote Sensing Change Detection

co-authors, Anita Madani — WACV 2026

DiffRegCD leverages diffusion models to predict dense flow fields between bi-temporal satellite images and warps multi-scale features before change prediction. The approach yields improved alignment in complex urban scenes while preserving genuine appearance changes and hard boundary details.

Selected Projects

WRIVA: Wide-Area Registration, Localization, and 3D Reconstruction

Johns Hopkins University / IARPA — 2024–Present

Contributing to the IARPA WRIVA program with research spanning image registration, visual localization, and large-scale 3D reconstruction. Developed correspondence and flow-based algorithms robust to extreme viewpoint shifts, doppelgänger scenes, and heterogeneous camera models across ground, UAV, and satellite imagery. Built multi-institution pipelines using Airflow, AWS, and Docker for automated evaluation and scalable benchmarking on real-world datasets.

Search Engine Optimization via Persian Semantic Graphs

2021–2022

Built a large-scale semantic graph over 2M+ Persian documents and applied GNN models (GCN, GAT) to obtain improved text embeddings. Increased retrieval quality using spectral clustering and PageRank-based ranking, reducing query latency by 23% and improving mean reciprocal rank (MRR) by 17%, enabling a significantly more responsive and accurate search engine pipeline.

Financial Deep Learning with Swarm Learning

2021

Implemented decentralized regression for financial time-series using swarm learning, combining LSTM and TCN architectures. Achieved a 7% reduction in RMSE while preserving institution-level privacy through federated-style weight sharing. Demonstrated the effectiveness of distributed learning under non-IID conditions.

Persian Speech Attribute Classification

Asr-e-Gooyesh Pardaz — Summer 2022

Constructed a 50k-utterance Persian speech dataset and trained deep neural networks for gender and age classification using TensorFlow and PyTorch. Reduced prediction error by 32% through dataset optimization, feature engineering, and architectural improvements specific to noisy, real-world speech signals.

Teaching

• Teaching Assistant, Medical Imaging Systems (EN.520.432), Johns Hopkins University, Fall 2025.
  Worked with Professor Jerry Prince. Designed homework and exam problems, prepared assignments and tests, and held weekly office hours and pre-exam review sessions focused on CT, MRI, SPECT, and PET imaging.
• Teaching Assistant, Image Analysis, Sharif University of Technology.
  Supported students with assignments and problem-solving on image processing, signals & systems, and medical imaging fundamentals.
• Teaching Assistant, Machine Learning Overview, Sharif University of Technology.
  Contributed to designing the final project and guided students through hands-on ML applications using Python and supervised/unsupervised learning methods.
• Teaching Assistant, Probability & Random Processes, Sharif University of Technology.
  Assisted with teaching probability theory, random variables, estimation, and MATLAB-based statistical analysis.
• Teaching Assistant, Analog Circuits & Lab, Sharif University of Technology.
  Taught transistor circuits, amplifiers, lab measurements, and simulation environments including SPICE and MATLAB.
• Teaching Assistant, Computer Architecture & Microprocessors, Sharif University of Technology.
  Guided students through assembly language, digital logic, CPU design, system architecture, and embedded systems fundamentals.
• Teaching Assistant, Logical Circuits & Digital Systems Lab, Sharif University of Technology.
  Taught combinational/sequential logic, Boolean algebra, Verilog, and digital circuit design with hands-on lab supervision.

Service

• Reviewer candidate, CVPR 2026.
• Led exam review and problem-solving sessions for Medical Imaging Systems (EN.520.432) at JHU.