Deep Object Co-Segmentation via Spatial-Semantic Network Modulation

Authors: Kaihua Zhang, Jin Chen, Bo Liu, Qingshan Liu12813-12820

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

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on four image co-segmentation benchmark datasets demonstrate the superior accuracy of the proposed method compared to state-of-the-art methods.
Researcher Affiliation Collaboration Kaihua Zhang,1 Jin Chen,1 Bo Liu,2 Qingshan Liu1 1B-DAT and CICAEET, Nanjing University of Information Science and Technology, Nanjing, China 2JD Finance America Corporation
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The codes are available at http://kaihuazhang.net/.
Open Datasets Yes We conduct extensive evaluations on four widely-used benchmark datasets (Faktor and Irani 2013; Rubinstein et al. 2013) including sub-set of MSRC, Internet, sub-set of i Coseg, and PASCAL-VOC. ... We adopt the COCO-SEG dataset released by (Wang et al. 2019) to train our model.
Dataset Splits No The paper mentions using several datasets for evaluation, including COCO-SEG for training, but does not explicitly state the specific train/validation/test splits (e.g., percentages or sample counts) used for these datasets within the paper.
Hardware Specification Yes Our model is implemented in Py Torch and a Nvidia RTX 2080Ti GPU is adopted for acceleration.
Software Dependencies No The paper states the model is 'implemented in Py Torch' but does not specify a version number for PyTorch or any other software dependencies.
Experiment Setup Yes We follow the same settings as (Wei et al. 2017; Wang et al. 2019): the input image group I consists of N = 5 images that are randomly selected from a group of images with co-object category, and a mini-batch of 4 I is fed into the model simultaneously during training. All images in I are resized to 224 224 as input... We leverage the Adam algorithm (Kingma and Ba 2014) to optimize the whole network in an end-to-end manner, among which the exponential decay rates for estimating the first and the second moments are set to 0.9 and 0.999, respectively. The learning rate starts from 1e-4 and reduces by a half every 25, 000 steps until the model converges at about 200,000 steps.