F³Net: Fusion, Feedback and Focus for Salient Object Detection

Authors: Jun Wei, Shuhui Wang, Qingming Huang12321-12328

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

Reproducibility Variable Result LLM Response
Research Type Experimental Comprehensive experiments on five benchmark datasets demonstrate that F3Net outperforms state-of-the-art approaches on six evaluation metrics. To demonstrate the performance of F3Net, we report experiment results on five popular SOD datasets and visualize some saliency maps. We conduct a series of ablation studies to evaluate the effect of each module.
Researcher Affiliation Academia 1Key Laboratory of Intelligent Information Processing of Chinese Academy of Sciences (CAS), Institute of Computing Technology, CAS, Beijing 100190, China 2University of Chinese Academy of Sciences, Beijing, 100049, China
Pseudocode Yes Algorithm 1: Cascaded Feedback Decoder
Open Source Code Yes Code will be released at https://github.com/weijun88/F3Net. Codes has been released.
Open Datasets Yes The performance of F3Net is evaluated on five popular datasets, including ECSSD (Yan et al. 2013) with 1000 images, PASCAL-S (Li et al. 2014) with 850 images, DUTOMRON (Yang et al. 2013) with 5168 images, HKU-IS (Li and Yu 2015) with 4,447 images and DUTS (Wang et al. 2017a) with 15,572 images. All datasets are human-labeled with pixel-wise ground-truth for quantitative evaluations.
Dataset Splits No The paper mentions DUTS-TR (training) and DUTS-TE (testing) splits but does not explicitly describe a separate validation split or its size/methodology.
Hardware Specification Yes An RTX 2080Ti GPU is used for acceleration.
Software Dependencies Yes We use Pytorch 1.3 to implement our model.
Experiment Setup Yes Maximum learning rate is set to 0.005 for Res Net-50 backbone and 0.05 for other parts. Warm-up and linear decay strategies are used to adjust the learning rate. The whole network is trained end-to-end, using stochastic gradient descent (SGD). Momentum and weight decay are set to 0.9 and 0.0005, respectively. Batchsize is set to 32 and maximum epoch is set to 32.