Learning Attention from Attention: Efficient Self-Refinement Transformer for Face Super-Resolution

Authors: Guanxin Li, Jingang Shi, Yuan Zong, Fei Wang, Tian Wang, Yihong Gong

IJCAI 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments demonstrate the effectiveness on a variety of datasets. The code is released at https://github.com/Guanxin-Li/LAA-Transformer.
Researcher Affiliation Academia 1School of Software Engineering, Xi an Jiaotong University 2Key Laboratory of Child Development and Learning Science, Southeast University 3Institute of Artificial Intelligence, Beihang University
Pseudocode No The paper describes its methods through text and diagrams (Figure 1, 2, 3) but does not include any explicit pseudocode or algorithm blocks.
Open Source Code Yes The code is released at https://github.com/Guanxin-Li/LAA-Transformer.
Open Datasets Yes The Celeb A [Liu et al., 2015] and the Helen [Le et al., 2012] are two publicly available face image datasets.
Dataset Splits No So we obtained about 178k images from Celeb A, of which 177k images were used as HR training images. In the testing phase, we extract the remaining 1000 images from the cropped Celeb A dataset and randomly extract 100 images from the cropped Helen dataset.
Hardware Specification Yes Our model is implemented in Py Torch and trained for 100 epochs using 2 NVIDIA Ge Force RTX 3090 GPUs.
Software Dependencies No Our model is implemented in Py Torch, but no specific version number for PyTorch or other software dependencies is provided.
Experiment Setup Yes The Adam W optimizer is used to train our model with β1 = 0.9, β2 = 0.99, and weight decay is set to 0.02. The learning rate is initially set to 4 10 4 and dropped by half every 20 epochs. Meanwhile, L1 loss is used for training. The batchsize is set to 32. Our model is implemented in Py Torch and trained for 100 epochs using 2 NVIDIA Ge Force RTX 3090 GPUs.