Robust Complex Behaviour Modeling at 90Hz

Authors: Xiangyu Kong, Yizhou Wang, Tao Xiang

AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Despite its simplicity, our experiments on three benchmark datasets show that it significantly outperforms the state-of-the-art for both temporal video segmentation and rare event detection.
Researcher Affiliation Academia Xiangyu Kong, Yizhou Wang Nat l Eng. Lab. for Video Technology Cooperative Medianet Innovation Center Key Lab. of Machine Perception (Mo E), School of EECS, Peking University, Beijing, 100871, China {kong, Yizhou.Wang}@pku.edu.cn Tao Xiang Queen Mary, University of London, London E1 4NS, United Kingdom t.xiang@qmul.ac.uk
Pseudocode No The paper describes the steps of its approach in prose and through equations but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not include any statements about providing open-source code for the described methodology or a link to a code repository.
Open Datasets Yes QMUL Junction Dataset (Hospedales, Gong, and Xiang 2012) ... MIT Traffic Dataset (Wang, Ma, and Grimson 2007) ... Subway Dataset (Hospedales, Gong, and Xiang 2012)
Dataset Splits No The paper specifies training and test sets but does not mention a separate validation set. For the QMUL and MIT datasets, the training set consists of normal clips from the first 40 minutes of the video, and the test set contains the remaining clips for rare event detection. For the Subway dataset, we used the first 5 minutes for training and the rest frames of the video for testing.
Hardware Specification No The paper mentions running on a "normal PC platform" and "ordinary PC" but does not specify any particular hardware details such as CPU, GPU models, or memory.
Software Dependencies No The paper mentions using "MATLAB" but does not provide a specific version number. No other software dependencies with version numbers are listed.
Experiment Setup Yes The grid cell size was set to 5 5 for QMUL and MIT, and 15 15 for Subway since the object sizes are larger in Subway. As the motion patterns of QMUL and MIT are more complex, the optical flow was quantized into No = 8 directions resulting in a Nb = 9 dimensional BCD descriptor. For the simpler motion patterns in Subway, we used a 5 dimensional BCD descriptor. The threshold Tb was set to 0.2 for all three datasets and its effect is analysed later. For computing the affinity matrix for scene decomposition, the radius R (see Eq. (1)) was set to 15 for QMUL and 8 for MIT, and 10 for Subway reflecting the scales of objects in each scene. ... In our experiments, we found that the refinement process converges after 2 3 iterations.