Anton Konushin

Samsung (Russia), Samsung (United States)

Papers

Total Citations

135

H-Index

About

Anton Konushin is a computer vision researcher whose work centers on 3D scene understanding, depth estimation, and perception systems for robotics and augmented reality. He is perhaps best known for **FCAF3D** (2022), a fully convolutional, anchor-free approach to 3D object detection that has garnered 115 citations, establishing him as a significant contributor to the field of point cloud-based detection. His research consistently tackles the challenge of enabling machines to robustly interpret three-dimensional environments from raw sensor data. Konushin's contributions span multiple facets of spatial perception. His work on **UniDet3D** addresses the critical limitation of small, domain-specific indoor datasets by training across multiple datasets simultaneously, improving generalization for real-world robotics and AR applications. His depth estimation research explores geometry-preserving models trained on diverse, uncalibrated stereo data mixtures — a practical approach to scaling single-view depth estimation. Earlier work, including the **DISCOMAN** dataset, provided the community with a valuable benchmark for semantic SLAM and navigation in simulated indoor environments. Collectively, Konushin's research reflects a sustained commitment to building scalable, generalizable 3D perception systems, making meaningful contributions to both foundational methods and the datasets that drive progress in embodied AI and spatial computing.