CRF-CNN: Modeling Structured Information in Human Pose Estimation

Authors: Xiao Chu, Wanli Ouyang, hongsheng Li, Xiaogang Wang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Its effectiveness is demonstrated through experiments on two benchmark datasets.
Researcher Affiliation Academia Xiao Chu The Chinese University of Hong Kong xchu@ee.cuhk.edu.hk Wanli Ouyang The Chinese University of Hong Kong wlouyang@ee.cuhk.edu.hk Hongsheng Li The Chinese University of Hong Kong hsli@ee.cuhk.edu.hk Xiaogang Wang The Chinese University of Hong Kong xgwang@ee.cuhk.edu.hk
Pseudocode Yes Algorithm 1 Message passing among features on factor graph.
Open Source Code No The paper does not provide any specific links or explicit statements about the availability of its source code.
Open Datasets Yes We conduct experiments on two benchmark datasets: the LSP dataset [12] and the FLIC dataset [18]... We also use negative samples, i.e. images not containing any person, from the INRIA dataset [5].
Dataset Splits No The paper specifies training and test splits for LSP ('1,000 images for training and 1,000 for test') and FLIC ('3,987 training images and 1,016 testing images'), but it does not mention a separate validation split or explicit cross-validation methodology.
Hardware Specification No The paper does not provide specific details about the hardware used for running experiments, such as CPU or GPU models, or memory specifications.
Software Dependencies No The paper mentions using CNNs, VGG, softmax loss, and standard BP, but it does not specify any software names with version numbers (e.g., PyTorch 1.x, TensorFlow 2.x).
Experiment Setup Yes During training, a whole image (or many of them) can be used as the mini-batch and the error at each output location of the network can be computed using an appropriate loss function with respect to the ground truth of the body joints. We use softmax loss with respect to the estimated part configuration z as the approximate loss function. ... End-to-end learning with softmax loss and standard BP is used. ... We set α 0.5 and β Nc, where Nc is the number of feature channels.