Edge-Aware Guidance Fusion Network for RGB–Thermal Scene Parsing

Authors: Wujie Zhou, Shaohua Dong, Caie Xu, Yaguan Qian3571-3579

AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments were performed on benchmark datasets to demonstrate the effectiveness of the proposed EGFNet and its superior performance compared with state-of-the-art methods. The code and results can be found at https://github.com/Shaohua Dong2021/EGFNet.
Researcher Affiliation Academia Wujie Zhou*, Shaohua Dong, Caie Xu, Yaguan Qian School of Information and Electronic Engineering, Zhejiang University of Science & Technology, Hangzhou, China. wujiezhou@163.com, shaohuadong2021@126.com, caiexu@163.com, qianyaguan@zust.edu.cn
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code Yes The code and results can be found at https://github.com/Shaohua Dong2021/EGFNet.
Open Datasets Yes We trained the EGFNet on the MFNet (Ha et al. 2017) and PST900 (Shivakumar et al. 2020) datasets.
Dataset Splits Yes We followed the training, testing, verification, and dataset splitting approaches used by (Ha et al. 2017). The PST900 dataset contains 894 aligned pairs of RGB and thermal images with pixel-level human annotations comprising five semantic classes, including the background. We used the splitting approach proposed by (Shivakumar et al. 2020) and resized each input image to 640 1280 pixels.
Hardware Specification Yes A computer equipped with an Intel 3.6 GHz i7 CPU and a single NVIDIA TITAN Xp graphics card was used for training and testing.
Software Dependencies Yes We used the Py Torch 1.7.0, CUDA 10.0, and cu DNN 7.6 libraries to implement the proposed EGFNet.
Experiment Setup Yes We trained EGFNet for 400 epochs and used the Ranger optimizer with an initial learning rate and weight decay of 5e-5 and 5e-4, respectively. We also used the weighted cross-entropy for both the semantic and boundary loss functions as well as weighting detailed by (Paszke et al. 2016).