Exploiting multimodal features has become a standard approach towards many video applications, including the video captioning task. One problem with the existing work is that it models the relevance of each type of features evenly, which neutralizes the impact of each individual modality to the word to be generated. In this paper, we propose a novel Modal Attention Network (MANet) to account for this issue. Our MANet extends the standard encoder-decoder network by adapting the attention mechanism to video modalities. As a result, MANet emphasizes the impact of each modality with respect to the word to be generated. Experimental results show that our MANet effectively utilizes multimodal features to generate better video descriptions. Especially, our MANet system was ranked among the top three systems at the 2nd Video to Language Challenge in both automatic metrics and human evaluations.

Automatic video description generation has recently been getting attention after rapid advancement in image caption generation. Automatically generating description for a video is more challenging than for an image due to its temporal dynamics of frames. Most of the work relied on Recurrent Neural Network (RNN) and recently attentional mechanisms have also been applied to make the model learn to focus on some frames of the video while generating each word in a describing sentence. In this paper, we focus on a sequence-to-sequence approach with temporal attention mechanism. We analyze and compare the results from different attention model configuration. By applying the temporal attention mechanism to the system, we can achieve a METEOR score of 0.310 on Microsoft Video Description dataset, which outperformed the state-of-the-art system so far.

Video event detection is a challenging problem in information and multimedia retrieval. Different from single action detection, event detection requires a richer level of semantic information from video. In order to overcome this challenge, existing solutions often represent videos using high level features such as concepts. However, concept-based representation can be confusing because it does not encode the relationship between concepts. This issue can be addressed by exploiting the co-occurrences of the concepts, however, it often leads to a very huge number of possible combinations. In this paper, we propose a new approach to obtain the relationship between concepts by exploiting the syntactic dependencies between words in the image captions. The main advantage of this approach is that it significantly reduces the number of informative combinations between concepts. We conduct extensive experiments to analyze the effectiveness of using the new dependency representation for event detection on two large-scale TRECVID Multimedia Event Detection 2013 and 2014 datasets. Experimental results show that i) Dependency features are more discriminative than concept-based features. ii) Dependency features can be combined with our current event detection system to further improve the performance. For instance, the relative improvement can be as far as 8.6% on the MEDTEST14 10Ex setting.

A complex event can be recognized by observing necessary evidences. In the real world scenarios, this is a difficult task because the evidences can happen anywhere in a video. A straightforward solution is to decompose the video into several segments and search for the evidences in each segment. This approach is based on the assumption that segment annotation can be assigned from its video label. However, this is a weak assumption because the importance of each segment is not considered. On the other hand, the importance of a segment to an event can be obtained by matching its detected concepts against the evidential description of that event. Leveraging this prior knowledge, we propose a new method, Event-driven Multiple Instance Learning (EDMIL), to learn the key evidences for event detection. We treat each segment as an instance and quantize the instance-event similarity into different levels of relatedness. Then the instance label is learned by jointly optimizing the instance classifier and its related level. The significant performance improvement on the TRECVID Multimedia Event Detection (MED) 2012 dataset proves the effectiveness of our approach.

A video can be viewed as a layered structure where the lowest layer are frames, the top layer is the entire video, and the middle layers are the sequences of consecutive frames or the concatenation of lower layers. While it is easy to find local discriminative features in video from lower layers, it is non-trivial to aggregate these features into a discriminative video representation. In literature, people often use sum pooling to obtain reasonable recognition performance on artificial videos. However, the sum pooling technique does not work well on complex videos because the region of interests may reside within some middle layers. In this paper, we leverage the layered structure of video to propose a new pooling method, named sum-max video pooling, to handle this problem. Basically, we apply sum pooling at the low layer representation while using max pooling at the high layer representation. Sum pooling is used to keep sufficient relevant features at the low layer, while max pooling is used to retrieve the most relevant features at the high layer, therefore it can discard irrelevant features in the final video representation. Experimental results on the TRECVID Multimedia Event Detection 2010 dataset shows the effectiveness of our method.