Label Error Correction and Generation through Label Relationships

Authors: Zijun Cui, Yong Zhang, Qiang Ji3693-3700

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental evaluations on six benchmark databases for two different tasks (facial action unit and object attribute classification) demonstrate the effectiveness of the proposed method in improving data annotation and in generating effective new labels.
Researcher Affiliation Collaboration Zijun Cui,1 Yong Zhang,2 Qiang Ji1 1Department of Electrical, Computer and Systems Engineering, Rensselaer Polytechnic Institute 1{cuiz3, jiq}@rpi.edu 2Tencent AI Lab 2zhangyong201303@gmail.com
Pseudocode No The paper describes methods in text and mathematical formulas but does not include any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not include any explicit statement about releasing source code for the described methodology, nor does it provide a link to a code repository.
Open Datasets Yes Datasets: For facial expressions, the Extended Cohn Kanande (CK+) (Lucey et al. 2010) database, the M&M Initiative facial expression database(Pantic et al. 2005)(MMI), BP4D-Spontaneous database(BP4D)(Zhang et al. 2013) and Emotion Net dataset(Matthews and Baker 2004) are four widely used databases for AU recognition.
Dataset Splits Yes To evaluate the performance, we performed 5 fold subject independent cross validation with F1-score as measurement. Each experiment was run 10 times, and the average F1-score was reported. ... For all experiments, the confidence level η is determined through a validation dataset.
Hardware Specification No The paper does not specify any particular hardware used for experiments, such as GPU models, CPU types, or memory.
Software Dependencies No The paper mentions software like 'regularized logistic regression model(LR)', 'SVM', and '3-layer CNN' but does not provide specific version numbers for any of these or other libraries/frameworks.
Experiment Setup Yes For all experiments, the confidence level η is determined through a validation dataset. ... Each classifier is trained with the improved labels and the original labels respectively. Then, the trained classifiers are used for AU prediction on the same testing set. Classifiers are trained to classify each AU independently. ... Each experiment was run 10 times, and the average F1-score was reported.