Dynamic Concept Composition for Zero-Example Event Detection
Authors: Xiaojun Chang, Yi Yang, Guodong Long, Chengqi Zhang, Alexander Hauptmann
AAAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | To validate the effectiveness of the proposed approach, we have conducted extensive experiments on the latest TRECVID MEDTest 2014, MEDTest 2013 and CCV dataset. The experimental results confirm the superiority of the proposed approach. |
| Researcher Affiliation | Academia | 1Centre for Quantum Computation and Intelligent Systems, University of Technology Sydney. 2Language Technologies Institute, Carnegie Mellon University. |
| Pseudocode | No | The paper describes the approach using mathematical formulations but does not include a distinct pseudocode block or algorithm section. |
| Open Source Code | No | The paper does not provide a direct link or explicit statement about releasing the source code for the proposed methodology. |
| Open Datasets | Yes | To evaluate the effectiveness of the proposed approach, we conduct extensive experiments on the following three large-scale event detection datasets: TRECVID MEDTest 2014 dataset (TRECVID 2014): This dataset has been introduced by the NIST... TRECVID MEDTest 2013 dataset (TRECVID 2013): The settings of MEDTest 2013 dataset is similar... Columbia Consumer Video dataset (Jiang et al. 2011): The official Columbia Consumer Video dataset... 3,135 concept detectors are pretrained using TRECVID SIN dataset (346 categories) (Over et al. 2014; Jiang et al. 2014), Google sports (478 categories) (Karpathy et al. 2014; Jiang et al. 2014), UCF101 dataset (101 categories) (Soomro, Zamir, and Shah 2012; Jiang et al. 2014), YFCC dataset (609 categories) (YFC ; Jiang et al. 2014) and DIY dataset (1601 categories) (Yu, Jiang, and Hauptmann 2014; Jiang et al. 2014). |
| Dataset Splits | No | The paper states using "official test split released by the NIST" but does not explicitly provide details about training/validation splits or their percentages for the main zero-exemplar event detection task. For few-exemplar, it mentions '10 Ex setting, where 10 positive videos are given for each event of interest' but no detailed split information. |
| Hardware Specification | No | The paper does not provide specific hardware details (e.g., CPU, GPU models, or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions several models and techniques (e.g., skip-gram model, Fisher vector, cascade SVM, improved dense trajectories with code of Wang and Schmid 2013) but does not provide specific software names with version numbers for implementation or dependencies. |
| Experiment Setup | Yes | In this paper, we empirically set k to 5. σ is the radius parameter of the Gaussian function. In our experiment, following (Liu et al. 2013; Lai et al. 2015) we set it as the mean value of all pairwise average distances among the videos. The dimension of each descriptor is first reduced by a factor of 2 and then use 256 components to generate the Fisher vectors. λ is a trade-off parameter among the two competing terms. |