Portrait of Mariia Gladkova

Mariia Gladkova

PhD Candidate, Technical University of Munich · Intern, Qualcomm AI Research

I am a PhD candidate at the Technical University of Munich (Chair for Computer Vision & Artificial Intelligence), advised by Prof. Daniel Cremers, and currently interning at Qualcomm AI Research in Amsterdam. My research spans visual SLAM and 3D reconstruction, spatial AI, and open-vocabulary 3D scene understanding. Before doing PhD, I completed Master's in Informatics at TUM (2017–2021), with my thesis on direct visual odometry with relocalization published at ICRA 2021, and Bachelor's in Computer Science at Constructor University (formerly Jacobs University Bremen, 2014–2017).

I am looking for full-time opportunities starting Summer 2027 — if you think I'd be a good fit for your team, please reach out!

  • mariia.gladkova [at] tum.de

News

Research

01

Visual SLAM & Place Recognition

Direct visual odometry with feature-based relocalization and cross-modal image–LiDAR place recognition for robust, large-scale localization.

02

3D Reconstruction, Tracking & Spatial AI

Multi-object tracking, sparse-view novel view synthesis, 4D scene understanding, and diagnostic benchmarking of open-vocabulary 3D detectors.

Selected Publications

For the full list, see my Google Scholar profile.

OV3D-Bench: A Diagnostic Benchmark for Open-Vocabulary Monocular 3D Detection

Mariia Gladkova, Neehar Peri, Ishan Khatri, Deva Ramanan, Daniel Cremers

ECCV Workshop (OpenSUN3D), 2026

A diagnostic benchmark that decouples localization, semantic robustness, and cross-domain transfer to reveal where open-vocabulary monocular 3D detectors actually fail.

arXiv · Code

VisDom: Sparse Novel View Synthesis with Visible Domain Constraint

Mariia Gladkova*, Tarun Yenamandra*, Edmond Boyer, Robert Maier, Tony Tung, Daniel Cremers (*equal contribution)

GCPR, 2026 (Oral)

Synthesizes novel views from sparse input images by constraining rendering to the visible domain, improving reconstruction quality when only a few views are available.

Project Page · arXiv · Code

TRASE: Tracking-free 4D Segmentation and Editing

Yun-Jin Li*, Mariia Gladkova*, Yan Xia, Daniel Cremers (*equal contribution)

3DV, 2026

Segments and edits dynamic 4D Gaussian-splatting scenes directly, without needing per-object tracking across frames.

Project Page · Code

VXP: Voxel-Cross-Pixel Large-scale Image-LiDAR Place Recognition

Yun-Jin Li*, Mariia Gladkova*, Yan Xia, Rui Wang, Daniel Cremers (*equal contribution)

3DV, 2025

Matches LiDAR point clouds and camera images at the voxel/pixel level for cross-modal place recognition at city scale.

Project Page · arXiv · Code

Gaussian Splatting in Style

Abhishek Saroha, Mariia Gladkova, Cecilia Curreli, Dominik Muhle, Tarun Yenamandra, Daniel Cremers

GCPR, 2024

Stylizes 3D Gaussian-splatting scenes in real time, generalizing to new style images without retraining per scene.

CASSPR: Cross Attention Single Scan Place Recognition

Yan Xia*, Mariia Gladkova*, Rui Wang, Qianyun Li, Uwe Stilla, João F. Henriques, Daniel Cremers (*equal contribution)

ICCV, 2023

Uses cross-attention between point and voxel representations to recover geometric detail lost during voxelization, improving single-scan LiDAR place recognition.

Code · arXiv

DirectTracker: 3D Multi-Object Tracking Using Direct Image Alignment and Photometric Bundle Adjustment

Mariia Gladkova, Nikita Korobov, Nikolaus Demmel, Aljoša Ošep, Laura Leal-Taixé, Daniel Cremers

IROS, 2022

Combines direct image alignment with sliding-window photometric bundle adjustment for accurate 3D multi-object tracking from monocular video.

Project Page · arXiv

Tight Integration of Feature-based Relocalization in Monocular Direct Visual Odometry

Mariia Gladkova, Rui Wang, Niclas Zeller, Daniel Cremers

ICRA, 2021

Integrates feature-based map relocalization into direct visual odometry at the camera tracking, bundle-adjustment, and pose-fusion stages for globally accurate camera poses.

Project Page · arXiv

Experience

Teaching & Service