AIDRC Seminar Series - Alexey Artemov

Title:

Geometric deep learning with the ABC dataset

Abstract:

ABC dataset is a collection of one million Computer-Aided Design (CAD) models for research of geometric deep learning methods and applications, providing explicitly parametrized curves and surfaces, exact differential quantities, patch segmentation, geometric feature detection, and shape reconstruction. Since its introduction in 2019, ABC dataset has enjoyed increasing popularity in the computer vision community. In this talk, we will consider an overview of the current development status with the ABC collection and discuss its use-cases within the recent published research, its possibility for extension and annotation. We will additionally present Deep Estimators of Features (DEFs), our newly proposed learning-based framework for predicting sharp geometric features in sampled 3D shapes, enabled by ABC data. With DEF, we take a step towards accurate 3D shape reconstruction; differently from existing data-driven methods, which reduce sharp geometric feature detection to feature classification, we propose to regress a scalar field representing the distance from point samples to the closest feature line on local patches. DEF is the first approach that scales to massive point clouds by fusing distance-to-feature estimates obtained on individual patches. We extensively evaluate our approach against related state-of-the-art methods on newly proposed synthetic and real-world 3D CAD model benchmarks. Our approach not only outperforms these (with improvements in Recall and False Positives Rates), but generalizes to real-world scans after training our model on synthetic data and fine-tuning it on a small dataset of scanned data.

Bio:

Alexey Artemov is a research associate at TUM Visual Computing Lab of Prof. Matthias Nießner. Previously to joining TUM, he was a researcher at CDISE Skoltech, advised by Associate Professor Evgeny Burnaev and Adjunct Professor Denis Zorin. Alexey obtained my Ph.D. in 2017 from Institute for Systems Analysis of Russian Academy of Sciences (ISA RAS) under the supervision of Prof. Evgeny Burnaev. Prior to joining Skoltech, he was a research engineer at Yandex, the Moscow-based internet giant, where he developed software systems for web search, computer vision, and autonomous driving.

Alexey’s research interests include 3D scene and shape reconstruction and digital geometry processing, where his work focuses on improving the geometry processing pipeline with deep learning accelerated algorithms. In 2021, Alexey received the Ilya Segalovich Award for academic advisors from Yandex.