Photo of Reza Heidari

Reza Heidari

M.Sc. Student · Aalto University

Espoo, Finland reza.heidari [at] aalto [dot] fi

Scholar GitHub LinkedIn

I am an M.Sc. student in Computer, Communication and Information Sciences at Aalto University, where I am writing my master's thesis in computer vision under the supervision of Prof. Juho Kannala. My thesis focuses on the automatic structural understanding of reconstructed 3D environments: recovering walls, wireframes and openings such as doors and windows from point clouds and images.

Before that, I was a research intern in the Visual AI Systems Research team at Nokia, working with Dr. Hamed R. Tavakoli on standardized neural network coding of implicit neural video representations and of visual tokens in vision–language models. Earlier, I worked on scene-graph-conditioned 3D scene generation with Dr. Federico Tombari at TUM, and on multimodal clinical prediction and medical visual question answering with Prof. Prasenjit Mitra at the L3S Research Center. I received my B.Sc. in Computer Engineering from Sharif University of Technology.

Research interests 3D computer vision & spatial understanding Multimodal learning Efficient visual & neural representations

Publications

arXiv2026

Compressing AI Traffic: Standardized Neural Network Coding of Visual-Token Representations in Split Vision–Language Inference

R. Heidari, H. R. Tavakoli, J. Kannala
arXiv preprint arXiv:2609.01200, 2026 Preprint
arXiv
MMSP2026

Neural Network Coding for Implicit Neural Video Representations

R. Heidari, H. R. Tavakoli, J. Kannala
IEEE International Workshop on Multimedia Signal Processing (MMSP), 2026 Accepted
IEEE Access2024

Robust Fusion of Time Series and Image Data for Improved Multimodal Clinical Prediction

A. Rasekh, R. Heidari, A. H. H. M. Rezaie, P. S. Sedeh, Z. Ahmadi, P. Mitra, W. Nejdl
IEEE Access, 2024 Published
arXiv
arXiv2026

Compressing AI Traffic: Standardized Neural Network Coding of Visual-Token Representations in Split Vision–Language Inference

R. Heidari, H. R. Tavakoli, J. Kannala
arXiv preprint arXiv:2609.01200, 2026 Preprint
arXiv
MMSP2026

Neural Network Coding for Implicit Neural Video Representations

R. Heidari, H. R. Tavakoli, J. Kannala
IEEE International Workshop on Multimedia Signal Processing (MMSP), 2026 Accepted
ADDR2025

Advances in Photoactivated Carbon-based Nanostructured Materials for Targeted Cancer Therapy

M. Eftekharifar, R. Heidari, N. Mohaghegh, A. Hassani, H. Heidari
Advanced Drug Delivery Reviews, 2025 Published
IEEE Access2024

Robust Fusion of Time Series and Image Data for Improved Multimodal Clinical Prediction

A. Rasekh, R. Heidari, A. H. H. M. Rezaie, P. S. Sedeh, Z. Ahmadi, P. Mitra, W. Nejdl
IEEE Access, 2024 Published
arXiv
In prep.

When Does Spatial Grounding Help VLMs? Exploring Region-Aware Visual Question Answering

R. Heidari, H. Anjidani, S. Yang, S. Luo, A. Rasekh, P. Mitra, J. Poon, C. Han
In preparation

Full list on Google Scholar.

Research Experience

  1. Aalto University · Computer Vision

    Sep 2026 – Present
    M.Sc. Thesis Researcher
    Supervisor: Prof. Juho Kannala
    • Researching automatic structural understanding of reconstructed 3D environments: recovering walls, wireframes and openings from point clouds and images.
    • Investigating a two-branch approach that combines point-cloud-based wall detection with image-based localization of doors and windows through cross-attention.
    3D reconstructionPoint cloudsCross-attention

  2. Nokia · Visual AI Systems Research

    Feb 2026 – Aug 2026
    Research Intern
    Supervisor: Dr. Hamed R. Tavakoli
    • Developed reproducible pipelines for standardized neural network coding (NNC) of implicit neural video representations and of intermediate visual features in vision–language models.
    • Evaluated Qwen3-VL video question answering with compressed visual representations, observing near-baseline accuracy with up to 98% reduction in transmitted BF16 tensor size.
    • Benchmarked NNC against fixed-bit quantization and arithmetic coding across NeRV-family architectures; NNC outperformed the baseline in 90% of matched-rate PSNR comparisons.
    Neural compressionImplicit neural representationsVision–language models

  3. Technical University of Munich · CAMP

    Jul 2024 – Nov 2024
    ML / Vision Research Intern
    Supervisor: Dr. Federico Tombari
    • Explored scene-graph-conditioned 3D scene generation with diffusion models, building on EchoScene.
    • Trained a task-specific variational autoencoder and investigated a unified graph diffusion approach for joint layout and shape generation, replacing separate generation branches.
    Diffusion models3D scene generationScene graphs

  4. L3S Research Center

    Jul 2023 – Nov 2024
    Machine Learning Intern (remote)
    Supervisor: Prof. Prasenjit Mitra
    • Developed multimodal clinical prediction models combining time series and chest X-rays through attention-based fusion and uncertainty-weighted multi-task learning (published in IEEE Access).
    • Contributed to medical visual question answering with collaborators at Penn State and in Australia, integrating entity extraction, SAM region proposals and PMC-CLIP alignment.
    Multimodal fusionMedical imagingVisual question answering

Teaching

Aalto University

2026
  • Seminar on Large Language ModelsAutumn 2026
    Teaching Assistant · Instructor: Eric Malmi.
  • Embedded Programming with Communication DevicesJan – May 2026
    Head Teaching Assistant · Co-developed a FreeRTOS-based C/C++ workspace and BLE support for Raspberry Pi Pico 2 W; supported students, grading and course coordination.
  • Differential and Integral Calculus 2Jan – Mar 2026
    Teaching Assistant · Led two exercise groups and contributed to grading.

Sharif University of Technology

2021 – 2025
  • Artificial Intelligence
    Lead Teaching Assistant
  • Data Structures, Probability and Statistics, Linear Algebra, Logic Design, Computer Architecture
    Teaching Assistant

Education

Aalto University

Aug 2025 – Dec 2026 (expected)
M.Sc., Computer, Communication and Information Sciences GPA 4.89 / 5.0
  • Thesis: structural understanding of reconstructed 3D environments (supervisor: Prof. Juho Kannala).
  • Selected coursework: Computer Vision, Deep Learning, Deep Generative Models, Reinforcement Learning, Visual Perception and Planning for Autonomous Driving, Supervised Machine Learning.

Sharif University of Technology

Sep 2021 – Jul 2025
B.Sc., Computer Engineering GPA 19.51 / 20.0
  • Thesis: rotation-invariant point-cloud classification using PointNet++.
  • Ranked 56 among ~126,000 participants in the National University Entrance Exam (top 0.05%).

Selected Projects

Speech Recognition and Post-ASR Correction

Oct 2025 – Dec 2025
Aalto University

ASR pipeline for modified Esperanto using Conformer-Transducer models, dictionary-based correction and ByT5 sequence-to-sequence refinement. Won the Aalto ASR course competition with roughly 2% word error rate.

PyTorchNVIDIA NeMoByT5

Hallucination Detection and Mitigation

2024
Code

Tennis 3D Analysis

Dec 2023 – Feb 2024

3D tennis video analysis pipeline combining player detection, ball trajectory tracking and pose classification.

PythonTensorFlowOpenCV Code

Awards