Davide Migliore
davidemigliore.bsky.social
Davide Migliore
@davidemigliore.bsky.social
Expert in neuromorphic vision | Technical business developer | Towards zero power, zero size, zero latency sensors | Tempo Sense
A method for Event-based Video Frame Interpolation (EVFI) using pre-trained video diffusion models to address the challenge of large motion between frames. This new approach outperforms existing methods and generalizes better across cameras
arxiv.org/pdf/2412.07761
#eventbased #arxiv
December 11, 2024 at 3:45 PM
A novel method for using event camera data in view synthesis via Gaussian Splatting! This approach leverages high temporal resolution and dynamic range to enhance reconstruction quality and speed, achieving better performance than existing NeRF methods.
arxiv.org/pdf/2412.07293
#arxiv #eventbased
December 11, 2024 at 3:41 PM
"Event Fields" leverage event cameras' high-speed, high-dynamic range for capturing light fields using innovative optical designs. Two frameworks—spatial and temporal multiplexing—enhancing scene understanding in extreme conditions.
arxiv.org/pdf/2412.06191
#eventbased #arxiv
December 10, 2024 at 10:37 PM
This paper proposes a framework to enhance images using event streams and locally poor-exposed frames, solving ghosting artifacts with dual-branch fusion. Results show improved image quality in extreme lighting.
openaccess.thecvf.com/content/ACCV...
#ACCV24 #eventbased
December 6, 2024 at 4:38 PM
High temporal resolution and dynamic range for robust high-speed any point tracking. A novel motion-robust feature alignment improves performance, showing a 135% better AJ metric and a 19% improvement on feature tracking benchmarks.
arxiv.org/pdf/2412.00133

#arxiv #eventbased
December 6, 2024 at 4:35 PM
New paper at #NEURIPS24
This paper introduces MOT principles to RGB-Event Single Object Tracking (SOT), using both RGB and event data to improve tracking robustness. Achieves state-of-the-art performance on benchmark datasets.
openreview.net/pdf/e2a89bd1...
#eventbased
December 5, 2024 at 2:14 PM
New paper at #NEURIPS24
Event-3DGS uses 3D Gaussian Splatting for event-based 3D reconstruction, solving NeRF training and editing challenges. It processes event data, reduces noise, and optimizes parameters for better quality.
github.com/lanpokn/Even...
#eventbased
December 5, 2024 at 2:00 PM
Combining event cameras with 3D Gaussian Splatting (3D-GS) for real-time dynamic scene reconstruction. #EventBased #ArxivPaper

arxiv.org/pdf/2411.16180
November 26, 2024 at 12:19 PM