Weigang Zhang

TL;DR: In this article, a new unconstrained UAV benchmark dataset is proposed for object detection, single object tracking, and multiple object tracking with new level challenges, including high density, small object, and camera motion, and a detailed quantitative study is performed using most recent state-of-the-art algorithms for each task.

TL;DR: In this article, a reinforcement learning-based method is proposed to select informative frame picking in video captioning, where the reward of each frame picking action is designed by maximizing visual diversity and minimizing discrepancy between generated caption and the ground truth.

TL;DR: In this article, a new unconstrained UAV benchmark is proposed for object detection, single object tracking, and multiple object tracking in complex scenarios with new level challenges, and the current state-of-the-art methods perform relative worse on the dataset, due to the new challenges appeared in UAV based real scenes, e.g., high density, small object, and camera motion.

TL;DR: A reinforcement-learning-based procedure to train the network sequentially, where the reward of each frame picking action is designed by maximizing visual diversity and minimizing textual discrepancy, so that a compact frame subset can be selected to represent the visual information and perform video captioning without performance degradation.

TL;DR: Deep Unsupervised Convolutional Domain Adaptation DUCDA method is proposed, which jointly minimizes the supervised classification loss of labeled source data and the unsupervised correlation alignment loss measured on both convolutional layers and fully connected layers.