PoseHD: Boosting Human Detectors Using Human Pose Information

Authors: Zhijian Liu, Bowen Pan, Yuliang Xiu, Cewu Lu

AAAI 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The experimental results on multiple pedestrian benchmark datasets validate that our proposed Pose HD framework can generally improve the overall performance of recent state-of-the-art human detectors (by 2-4% in both m AP and MR metrics).
Researcher Affiliation Academia Shanghai Jiao Tong University, Shanghai, China 200240
Pseudocode No The paper does not contain pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide an explicit statement or link for open-source code for its methodology.
Open Datasets Yes The performance of Pose HD is evaluated across three pedestrian benchmark datasets: INRIA (Dalal and Triggs 2005), ETH (Ess, Leibe, and Van Gool 2007) and PASCAL VOC 2007 (Everingham et al. 2010).
Dataset Splits No The paper mentions training and testing on datasets but does not provide specific training/validation/test dataset splits (e.g., percentages or counts).
Hardware Specification Yes We use Tensor Flow (Abadi et al. 2016) and Nvidia Titan X GPU to train the classification network.
Software Dependencies No The paper mentions software like Tensor Flow, Adam, and Lambda MART, but does not provide specific version numbers for these dependencies.
Experiment Setup Yes We initialize the weights randomly from scratch, and the optimization is carried out using Adam (Kingma and Ba 2015) with β1 = 0.9 and β2 = 0.999. We use a fixed learning rate of γ = 10 3 and mini-batch size of 128.