CoADNet: Collaborative Aggregation-and-Distribution Networks for Co-Salient Object Detection

Authors: Qijian Zhang, Runmin Cong, Junhui Hou, Chongyi Li, Yao Zhao

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

Reproducibility Variable Result LLM Response
Research Type Experimental The proposed Co ADNet is evaluated on four prevailing Co SOD benchmark datasets, which demonstrates the remarkable performance improvement over ten state-of-the-art competitors.
Researcher Affiliation Academia 1Department of Computer Science, City University of Hong Kong, Hong Kong SAR, China 2Institute of Information Science, Beijing Jiaotong University, China 3Beijing Key Laboratory of Advanced Information Science and Network Technology, China 4School of Computer Science and Engineering, Nanyang Technological University, Singapore
Pseudocode No The paper contains architectural diagrams (e.g., flowcharts) but no clearly labeled pseudocode or algorithm blocks.
Open Source Code Yes https://rmcong.github.io/proj_Co ADNet.html and We also provide a Pytorch implementation version
Open Datasets Yes In experiments, we conduct extensive evaluations on four popular datasets, including Co SOD3k [15], Cosal2015 [53], MSRC [47], and i Coseg [2]. and In each training iteration, 24 sub-groups from the co-saliency dataset COCO-SEG [43] and 64 samples from the saliency dataset DUTS [44] are simultaneously fed into the network
Dataset Splits No The paper mentions datasets used for evaluation (test sets) and training, but does not explicitly describe a separate validation set or split for model tuning.
Hardware Specification Yes The proposed framework is implemented in Mind Spore and accelerated by 4 Tesla P100 GPUs 3.
Software Dependencies No The proposed framework is implemented in Mind Spore and accelerated by 4 Tesla P100 GPUs 3. We also provide a Pytorch implementation version. (No specific version numbers are provided for Mind Spore or Pytorch).
Experiment Setup Yes In each training iteration, 24 sub-groups from the co-saliency dataset COCO-SEG [43] and 64 samples from the saliency dataset DUTS [44] are simultaneously fed into the network for jointly optimizing the objective function in Eq. 9, where α = 0.7 and β = 0.3, by the Adam [29] algorithm with a weight decay of 5e 4. and we set the initial learning rate to 1e 4 that is halved every 5, 000 iterations, and the whole training process converges until 50, 000 iterations.