Source author record

Siwen Wang

Siwen Wang appears in the imported research catalog. Authorship, coauthor and topic links are available while profile ownership is still unclaimed.

ResearcherUnclaimed source record

Computer Vision hep-ph physics.ao-ph physics.geo-ph physics.soc-ph Social and Information Networks

Catalog footprint

What is connected

4works

6topics

4close collaborators

Actions

Connect this record

Open graph Browse works

Inspect adjacent papers, topics, institutions and collaborators without losing the researcher page.

Building this map preview

BZPEER is loading the nearby papers, people, topics and institutions for this page.

preprint2020arXiv

Object Instance Mining for Weakly Supervised Object Detection

Weakly supervised object detection (WSOD) using only image-level annotations has attracted growing attention over the past few years. Existing approaches using multiple instance learning easily fall into local optima, because such mechanism tends to learn from the most discriminative object in an image for each category. Therefore, these methods suffer from missing object instances which degrade the performance of WSOD. To address this problem, this paper introduces an end-to-end object instance mining (OIM) framework for weakly supervised object detection. OIM attempts to detect all possible object instances existing in each image by introducing information propagation on the spatial and appearance graphs, without any additional annotations. During the iterative learning process, the less discriminative object instances from the same class can be gradually detected and utilized for training. In addition, we design an object instance reweighted loss to learn larger portion of each object instance to further improve the performance. The experimental results on two publicly available databases, VOC 2007 and 2012, demonstrate the efficacy of proposed approach.

preprint2020arXiv

Spring Festival points the way to cleaner air in China

Human migration during the Chinese Spring Festival (SF) is the largest collective human activity of its kind in the modern era-involving about one-tenth of the world population and over six percent of the earth's land surface area. The festival results in a drop of air pollutant emissions that causes dramatic changes of atmospheric composition over China's most polluted regions. Based on satellite and in-situ measurements for the years 2005-2019 over 50 cities in eastern China, we find that the atmospheric NO2 pollution dropped by ~40% during the SF week, and fine particulate matter (PM2.5) decreased by ~30% in the following week, reflecting the effectiveness of precursor emission controls on the mitigation of secondary PM2.5 formation. However, although human activity and emissions are at the lowest level, air pollution over eastern China during the SF still far exceeds that over other worldwide pollution hotspots. Our analyses suggest that measures based solely on end-of-pipe controls and industry upgrades may not suffice to meet air quality goals. Further cleaning of the air in China depends fundamentally on sustainable advances in both heavy industry upgrades and clean energy transition.

preprint2015arXiv

The effect of social welfare system based on the complex network

With the passage of time, the development of communication technology and transportation broke the isolation among people. Relationship tends to be complicated, pluralism, dynamism. In the network where interpersonal relationship and evolved complex net based on game theory work serve respectively as foundation architecture and theoretical model, with the combination of game theory and regard public welfare as influencing factor, we artificially initialize that closed network system. Through continual loop operation of the program, we summarize the changing rule of the cooperative behavior in the interpersonal relationship, so that we can analyze the policies about welfare system about whole network and the relationship of frequency of betrayal in cooperative behavior. Most analytical data come from some simple investigations and some estimates based on internet and environment and the study put emphasis on simulating social network and analyze influence of social welfare system on Cooperative Behavio.

preprint1994arXiv

Pion Form Factor: Transition From Soft To Hard QCD

We have reexamined the elastic pion form factor over a broad range of momentum transfers with the mass evolution from current quark to constituent quark being taken into account. We have also studied the effect of Sudakov form factors, of anomalous quark magnetic moments and of alternative soft wave functions. Our calculation shows a power-law falloff from the present experimental values to near-asymptotic values in the few $GeV^{2}$ range.