This paper proposes a visual scene decomposition framework that applies geometry-aware object clustering and temporal smoothing for robust scene interpretation in autonomous navigation systems across varying environments.

Daniel Joseph Whitman, Zhang Xiaoli, Rachita Sandeep Kulkarni, Antoine Pierre Laurent, Naomi Erika Fujimura, Lucia Helena Paredes

This study presents a model for egocentric action understanding based on hand-object interaction graphs and temporal semantic encoding, enabling accurate recognition of fine-grained tasks from first-person video streams.

Samuel Edward Linton, Liu Fangyi, Kavya Pranav Menon, Louis Henri Dubois, Yuka Noriko Yamashita, Ana Celeste Moreira

This paper introduces a drone surveillance model leveraging vision transformers and scale-invariant density estimation, enabling accurate real-time crowd monitoring during public events from high-altitude perspectives.

Christopher Allan Peters, Zhang Yunhua, Shruthi Anuja Balan, Matteo Rinaldi Romano, Hana Miyako Sakamoto, Beatriz Luz Ramírez

This study proposes a neural implicit representation method for scene completion in robotic systems, using sparse depth input and semantic priors to reconstruct occluded regions in complex indoor environments.

Gregory Isaac Campbell, Huang Wenxiu, Anitha Ramesh Iyer, André Louis Marchand, Emi Yurika Takeda, Daniela Rosario Herrera

This research proposes an adversarial domain adaptation technique for vehicle detection across cities, employing multi-scale pseudo-labeling and consistency constraints to generalize models trained on limited regional data.

Julian Frederick Hargrove, Zhang Lingxi, Anuja Rajan Bhatia, Jean-Claude René Fournier, Haruka Natsuki Nakamoto, Lorena Elisa Castillo

This study introduces a real-time object counting system for manufacturing environments using vision transformers and temporal voting, improving throughput monitoring and reducing errors in industrial automation workflows.

Colin Raymond Foster, Zhang Lianmei, Rupal Nivedita Borkar, Antoine Gabriel Morel, Yumi Sachiko Tanabe, Carolina Beatriz Fuentes

This paper proposes a self-evolving visual inspection framework using anomaly-aware transformers and continuous learning, enabling adaptive defect detection across dynamic production environments and material types.

Isaiah Douglas Trent, Liu Meiqing, Nisha Preeti Joshi, Baptiste Arnaud Lefebvre, Ayumi Haruka Matsumoto, Mariana Lucia Andrade