Results 1 -
3 of
3
Semantic Concept Discovery for Large-Scale Zero-Shot Event Detection
"... We focus on detecting complex events in uncon-strained Internet videos. While most existing works rely on the abundance of labeled training data, we consider a more difficult zero-shot setting where no training data is supplied. We first pre-train a number of concept classifiers using data from othe ..."
Abstract
-
Cited by 1 (1 self)
- Add to MetaCart
We focus on detecting complex events in uncon-strained Internet videos. While most existing works rely on the abundance of labeled training data, we consider a more difficult zero-shot setting where no training data is supplied. We first pre-train a number of concept classifiers using data from other sources. Then we evaluate the semantic correla-tion of each concept w.r.t. the event of interest. Af-ter further refinement to take prediction inaccuracy and discriminative power into account, we apply the discovered concept classifiers on all test videos and obtain multiple score vectors. These distinct score vectors are converted into pairwise compari-son matrices and the nuclear norm rank aggregation framework is adopted to seek consensus. To ad-dress the challenging optimization formulation, we propose an efficient, highly scalable algorithm that is an order of magnitude faster than existing alter-natives. Experiments on recent TRECVID datasets verify the superiority of the proposed approach. 1
Encoding Feature Maps of CNNs for Action Recognition
"... Abstract We describe our approach for action classification in the THUMOS Challenge 2015. Our approach is based on two types of features, improved dense trajectories and CNN features. For trajectory features, we extract HOG, HOF, MBHx, and MBHy descriptors and apply Fisher vector encoding. For CNN ..."
Abstract
- Add to MetaCart
(Show Context)
Abstract We describe our approach for action classification in the THUMOS Challenge 2015. Our approach is based on two types of features, improved dense trajectories and CNN features. For trajectory features, we extract HOG, HOF, MBHx, and MBHy descriptors and apply Fisher vector encoding. For CNN features, we utilize a recent deep CNN model, VGG19, to capture appearance features and use VLAD encoding to encode/pool convolutional feature maps which shows better performance than average pooling of feature maps and full-connected activation features. After concatenating them, we train a linear SVM classifier for each class in a one-vs-all scheme.
Fisher Kernel Temporal Variation-based Relevance Feedback for Video Retrieval
"... This paper proposes a novel framework for Relevance Feedback based on the Fisher Kernel (FK). Specifically, we train a Gaussian Mixture Model (GMM) on the top retrieval results (without supervision) and use this to create a FK representation, which is therefore specialized in modelling the most rele ..."
Abstract
- Add to MetaCart
(Show Context)
This paper proposes a novel framework for Relevance Feedback based on the Fisher Kernel (FK). Specifically, we train a Gaussian Mixture Model (GMM) on the top retrieval results (without supervision) and use this to create a FK representation, which is therefore specialized in modelling the most relevant examples. We use the FK representation to explicitly capture temporal variation in video via frame-based features taken at different time intervals. While the GMM is being trained, a user selects from the top examples those which he is looking for. This feedback is used to train a Support Vector Machine on the FK representation, which is then applied to re-rank the top retrieved results. We show that our approach outperforms other state-of-the-art relevance feedback methods. Experiments were carried out on the Blip10000, UCF50, UCF101 and ADL standard datasets using a broad range of multi-modal content descriptors (visual, audio, and text).
