Background Suppression Network for Weakly-Supervised Temporal Action Localization

Authors: Pilhyeon Lee, Youngjung Uh, Hyeran Byun11320-11327

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness of Ba S-Net and its superiority over the state-of-the-art methods on the most popular benchmarks THUMOS 14 and Activity Net.
Researcher Affiliation Collaboration Pilhyeon Lee Yonsei University lph1114@yonsei.ac.kr Youngjung Uh Clova AI Research, NAVER Corp. youngjung.uh@navercorp.com Hyeran Byun Yonsei University hrbyun@yonsei.ac.kr
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes Our code and the trained model are available at https://github.com/Pilhyeon/Ba SNet-pytorch.
Open Datasets Yes We conduct experiments on weakly-supervised temporal action localization task on the most popular benchmarks: THUMOS 14 (Jiang et al. 2014) and Activity Net (Caba Heilbron et al. 2015).
Dataset Splits Yes We conduct experiments on weakly-supervised temporal action localization task on the most popular benchmarks: THUMOS 14 (Jiang et al. 2014) and Activity Net (Caba Heilbron et al. 2015). We also evaluate our Ba S-Net on Activity Net1.3 in Table 3. Experimental results on Activity Net1.2 are shown in Table 4.
Hardware Specification Yes Experiments are conducted on a single GTX 1080Ti GPU.
Software Dependencies No The paper mentions using 'pre-trained feature extractor' and 'TVL1 algorithm' but does not provide specific version numbers for any software dependencies.
Experiment Setup Yes All hyperparameters are empirically determined by grid search; r = 8, α = 1, β = 1, γ = 10 4, and θclass = 0.25. For θact, we use a set of thresholds from 0 to 0.5 with the step 0.025 and perform non-maximum suppression (NMS) with threshold 0.7 to remove highly overlapped proposals.