We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,670 hours of dailylife activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 931 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards, with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception.

Small object detection is an interesting topic in computer vision. With the rapid development in deep learning, it has drawn attention of several researchers with innovations in approaches to join a race. These innovations proposed comprise region proposals, divided grid cell, multiscale feature maps, and new loss function. As a result, performance of object detection has recently had significant improvements. However, most of the state-of-the-art detectors, both in one-stage and two-stage approaches, have struggled with detecting small objects. In this study, we evaluate current state-of-the-art models based on deep learning in both approaches such as Fast RCNN, Faster RCNN, RetinaNet, and YOLOv3. We provide a profound assessment of the advantages and limitations of models. Specifically, we run models with different backbones on different datasets with multiscale objects to find out what types of objects are suitable for each model along with backbones. Extensive empirical evaluation was conducted on 2 standard datasets, namely, a small object dataset and a filtered dataset from PASCAL VOC 2007. Finally, comparative results and analyses are then presented.

Abstract. Author name ambiguity is one of the problems that decrease the quality and reliability of information retrieved from digital libraries. Existing methods have tried to solve this problem by predefining a feature set based on expert's knowledge for a specific dataset. In this paper, we propose a new approach which uses deep neural network to learn features automatically for solving author name ambiguity. Additionally, we propose the general system architecture for author name disambiguation on any dataset. We evaluate the proposed method on a dataset containing Vietnamese author names. The results show that this method significantly outperforms other methods that use predefined feature set. The proposed method achieves 99.31% in terms of accuracy. Prediction error rate decreases from 1.83% to 0.69%, i.e., it decreases by 1.14%, or 62.3% relatively compared with other methods that use predefined feature set (Table 3).

No abstract

No abstract

We introduce Ego4D, a massive-scale egocentric video dataset and benchmark suite. It offers 3,025 hours of dailylife activity video spanning hundreds of scenarios (household, outdoor, workplace, leisure, etc.) captured by 855 unique camera wearers from 74 worldwide locations and 9 different countries. The approach to collection is designed to uphold rigorous privacy and ethics standards with consenting participants and robust de-identification procedures where relevant. Ego4D dramatically expands the volume of diverse egocentric video footage publicly available to the research community. Portions of the video are accompanied by audio, 3D meshes of the environment, eye gaze, stereo, and/or synchronized videos from multiple egocentric cameras at the same event. Furthermore, we present a host of new benchmark challenges centered around understanding the first-person visual experience in the past (querying an episodic memory), present (analyzing hand-object manipulation, audio-visual conversation, and social interactions), and future (forecasting activities). By publicly sharing this massive annotated dataset and benchmark suite, we aim to push the frontier of first-person perception.

No abstract

This paper addresses the problem of autonomous quadrotor navigation within indoor spaces. In particular, we focus on the case where a visual map of the area, represented as a graph of linked images, is constructed offline (from visual and potentially inertial data collected beforehand) and used to determine visual paths for the quadrotor to follow. In addition, during the actual navigation, the quadrotor employs both wide-and short-baseline random sample consensuses (RANSACs) to efficiently determine its desired motion toward the next reference image and handle special motions, such as rotations in place. In particular, when the quadrotor relies only on visual observations, it uses the 5pt and 2pt algorithms in the wide-and short-baseline RANSACs, respectively. On the other hand, when information about the gravity direction is available, significant gains in speed are realized by using the 3pt + 1 and 1pt + 1 algorithms instead. Lastly, we introduce an adaptive optical-flow algorithm that can accurately estimate the quadrotor's horizontal velocity under adverse conditions (e.g., when flying over dark, textureless floors) by progressively using information from more parts of the images. The speed and robustness of our algorithms are evaluated experimentally using a commercial-off-the-shelf quadrotor navigating in the presence of dynamic obstacles (i.e., people walking) along lengthy corridors and through tight corners, as well as across building floors via poorly lit staircases.