Co-attention CNNs for Unsupervised Object Co-segmentation

Authors: Kuang-Jui Hsu, Yen-Yu Lin, Yung-Yu Chuang

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

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments show that our method achieves superior results, even outperforming the state-of-the-art, supervised methods.
Researcher Affiliation Academia 1Academia Sinica, Taiwan 2National Taiwan University, Taiwan
Pseudocode No The paper describes the proposed method and optimization process in text and equations but does not include any pseudocode or clearly labeled algorithm blocks.
Open Source Code No The paper does not provide any specific links to source code repositories or explicitly state that the code for the described methodology is publicly available in supplementary material or elsewhere.
Open Datasets Yes Our method is evaluated on three benchmarks for co-segmentation, the Internet dataset [Rubinstein et al., 2013], the i Coseg dataset [Batra et al., 2010], and the PASCAL-VOC dataset [Faktor and Irani, 2013].
Dataset Splits No The paper mentions using specific datasets for evaluation but does not provide explicit details about training, validation, or test dataset splits (e.g., specific percentages or sample counts for each split). The evaluation is on entire benchmark datasets.
Hardware Specification Yes model training and object mask refinement in each of 60 epochs take about 20 and 6 seconds, respectively, on an NVIDIA Titan X graphics card.
Software Dependencies No The paper mentions several software components like 'Mat Conv Net', 'Res Net50', 'VGG-16', 'FCN', 'ADAM', and 'geodesic object proposal (GOP)'. However, it does not provide specific version numbers for these software dependencies, which is crucial for reproducibility.
Experiment Setup Yes The learning rate is set to 10^-6 and kept fixed during optimization...The total number of epoches is 60. The batch size, weight decay, and momentum are set to 5, 0.0005, and 0.9, respectively. All images for co-segmentation are resized to the resolution 384x384 in advance...The parameter λ in Eq. (1) is empirically set and fixed to 9 in all experiments.