Low-Light Face Super-resolution via Illumination, Structure, and Texture Associated Representation

Authors: Chenyang Wang, Junjun Jiang, Kui Jiang, Xianming Liu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments demonstrate that the proposed joint optimization framework achieves significant improvements in reconstruction quality and perceptual quality over existing two-stage sequential solutions.
Researcher Affiliation Academia Chenyang Wang, Junjun Jiang*, Kui Jiang, Xianming Liu School of Computer Science and Technology, Harbin Institute of Technology, Harbin, China {wangchy02, jiangjunjun, jiangkui, csxm}@hit.edu.cn
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code Yes Code is available at https://github.com/wcy-cs/IC-FSRDENet.
Open Datasets Yes We conduct experiments on commonly-used high quality face dataset Celeb AMask-HQ (Yu et al. 2018).
Dataset Splits Yes From it, we randomly choose 3050 face images for training, and 600 for validation, and another 300 for testing.
Hardware Specification Yes The experiments are implemented on Py Torch with one NVIDIA 3090 GPU.
Software Dependencies No The experiments are implemented on Py Torch with one NVIDIA 3090 GPU. This only mentions “Py Torch” without a specific version number and does not list other software dependencies with versions.
Experiment Setup Yes In IC-FSRNet, L is set as 14. IC-FSRNet and DENet are trained successively and individually. The optimizer is Adam with β1=0.9, β2=0.99, and ϵ=1e-8. The learning rate is 1e-4 for both two networks. Towards the DENet, the backbone is UNet and other settings follow (Saharia et al. 2021).