Iaro Melekhov

I am a research scientist at Sharper Shape and a postdoc (part-time) at Aalto University in Juho Kannala's Research Group My research interests lie at the intersection of 3D computer vision, computer graphics, and machine learning. I am particularly interested in scene understanding, camera relocalization and mapping, and developing new machine learning methods for 3D computer vision applications.

Previously, I was a research intern at Niantic Labs (London) under the supervision of Prof. Gabriel Brostow and Dr. Daniyar Turmukhambetov. Before this, I spent 6 months in Cambridge, UK working on visual perception algorithms for autonomous cars and having fun at Wayve. I also did a research visit to Marc Pollefeys' Computer Vision and Geometry Group in the Department of Computer Science at ETH Zurich.

Email  /  CV  /  Google Scholar  /  X  /  Github

photo taken by Alena Pugacheva

Research

My research interests include machine learning for 3D computer vision, generative AI, large-scale visual mapping and localization, end-to-end learning and feature matching with applications in autonomous driving, AR/VR.

News

Publications

The full list of my papers can be found on Google Scholar. The list of selected projects:

Differentiable Product Quantization for Memory Efficient Camera Relocalization
Zakaria Laskar*, Iaroslav Melekhov*, Assia Benbihi, Shuzhe Wang, Juho Kannala
ECCV, 2024
paper / project page / code

We propose a differentiable product quantization layer to address the memory efficiency of the camera relocalization pipeline.

DN-Splatter: Depth and Normal Priors for Gaussian Splatting and Meshing
Matias Turkulainen*, Xuqian Ren*, Iaroslav Melekhov, Otto Seiskari, Esa Rahtu, Juho Kannala
arXiv, 2024
paper / project page / code

We extend 3D Gaussian splatting with depth and normal cues to tackle challenging indoor datasets and showcase techniques for efficient mesh extraction.

ECLAIR: A High-Fidelity Aerial LiDAR Dataset for Semantic Segmentation
Iaroslav Melekhov*, Anand Umashankar*, Hyeong-Jin Kim, Vladislav Serkov, Dusty Argyle
CVPR Workshops, 2024
paper / project page / code

We introduce ECLAIR, a diverse and high-fidelity aerial LiDAR dataset for point cloud semantic segmentation.

Digging Into Self-Supervised Learning of Feature Descriptors
Iaroslav Melekhov*, Zakaria Laskar*, Xiaotian Li, Shuzhe Wang, Juho Kannala
3DV, 2021
paper / project page / code

We show how to use unsupervised learning to optimize a CNN-based local descriptor that is robust to illumination changes and competitve with its fully-supervised counterparts.


This website is based on the source code of Jon Barron's and Torsten Sattler's websites.
Profile picture taken by .