Hybrid Sharing for Multi-Label Image Classification

Authors: Zihao Yin, Chen Gan, Kelei He, Yang Gao, Junfeng Zhang

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments are conducted on two benchmark datasets, with the results demonstrating that the proposed method achieves state-of-the-art performance and yields simultaneous improvements across most labels.
Researcher Affiliation Academia Zihao Yin2,3, Chen Gan2,3, Kelei He1,3 , Yang Gao2,3 and Junfeng Zhang1,3 1 Medical School of Nanjing University 2 State Key Laboratory for Novel Software Technology, Nanjing University 3 National Institute of Healthcare Data Science, Nanjing University {zihao.yin, chengan}@smail.nju.edu.cn {hkl, gaoy, jfzhang}@nju.edu.cn
Pseudocode Yes Algorithm 1 outlines the detailed mechanism of the Mo E mechanism that we applied.
Open Source Code No The code is available at this URL. (Abstract) / The code will be available upon acceptance. (Reproducibility Statement)
Open Datasets Yes We have performed extensive experiments on two datasets, MS-COCO and PASCAL VOC, to verify the superiority of our model. [...] MS-COCO (Lin et al., 2014) is a large dataset of 80 object classes [...] PASCAL-VOC (Everingham et al., 2015) 2007 is also a well-acknowledged dataset for multi-label image classification.
Dataset Splits Yes We have performed extensive experiments on two datasets, MS-COCO and PASCAL VOC...We follow previous work to train on train-val set and validate on test set on 2007 version.
Hardware Specification Yes All experiments are run on 4 Tesla V100-SXM2-32GB.
Software Dependencies No The paper mentions 'timm' and 'Adam optimizer' and implies Python usage but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes Unless otherwise stated, the following setting is valid for all experiments. We resize all input images from any dataset to Hi Wi = 576 576... We train the model for 100 epochs using the Adam optimizer, with weight decay of 1e-2, (β1, β2) = (0.9, 0.9999), and a learning rate of 1e-4.