Abstract
Video concept learning often requires a large set of train-ing samples.In practice,however;acquiring noise -free training labels with suficient positive examples is very ex-pensive.A plausible solution for training data collection is by sampling from the vast quantities of images and videos on the Web.Such a solution is morivated by the assumption that the retrieved images or videos are highly correlated with the query.Still,a nuimber of challenges remain.First,Web videos are often untrimmed.Thus,only parts of the videos are relevant to the query,Second,the retrieved Web images are always highly relevant to the issued query.How-ever,thoughtlessly utilizing the images in the video domain may even hurt the performance due to the well-known se-mantic drift and domain gap problems.As a result,a valid question is how Web images and videos interact for video concept learning.In this paperwe propose a Lead-Exceed Neural Network(LENN),which reinforces the training on Web images and videos in a curriculum mannerSpecif ically,the training proceeds by inputting frames of Web videos to obtain a network.The Web images are then fil-tered by the learnt nerwork and the selected images are ad-ditionally fed into the network to enhance the architecture and further trim the videos,In addition,Long Short-Term Memory(LSTM)can be applied on the trimmed videos to explore temporal information.Encouraging results are re-ported on UCF1O1,TRECVID 2013 and 2014 MEDTest in the context of both action recognition and event detection.Without using human annotated exemplars,our proposed LENN can achieve 74.4%accuracy on UCFI01 dataset.