INSAIT participated with 5 accepted papers at the International Conference on Computer Vision – ICCV, a leading conference in computer vision, taking place from 2nd to 6th October in Paris. For the first time in the 36-year history of ICCV, it is held with the participation of a Bulgarian institution through INSAIT. This success came shortly after INSAIT had 7 papers accepted at NeurIPS 2023, a top conference in machine learning. The ICCV papers include general topics such as autonomous driving, surveillance and security, virtual reality, human image modelling and others. More details below on each paper:
Source-free Depth for Object Pop-out (finding hidden objects)
This paper proposes a new method for detecting masked objects that are difficult to identify by humans, with applications in surveillance and security.
Deformable Neural Radiance Fields using RGB and Event Cameras
The paper proposes a new method for capturing complex dynamic shapes, such as fire and water, using biotechnology cameras, with applications in a number of fields, such as virtual reality.
Improving Online Lane Graph Extraction by Object-Lane Clustering
The paper presents a method that helps self-driving cars better understand their surroundings. It uses 3D data to improve the car’s ability to recognize roads and objects, making it safer and more reliable for autonomous driving.
Surface Normal Clustering for Implicit Representation of Manhattan Scenes
This paper proposes a new method for creating 3D models and images of buildings and enclosed spaces.
PATMAT: Person Aware Tuning of Mask-Aware Transformer for Face inpainting
This paper proposes a novel face imaging method that effectively preserves identity and improves existing models in terms of image quality and face-specific details.
All publications can be found at: https://insait.ai/publications/