BGAI

Broadcast-Gated Attention with Identity Adaptive Integration for Efficient Image Super-Resolution

Qian Wang, Yanyu Mao, Ruilong Guo, Mengyang Wang, Jing Wei, Han Pan

💻Environment

• PyTorch >= 1.9
• Python 3.7
• Numpy
• BasicSR >= 1.3.4.9

🔧Installation

pip install -r requirements.txt

📜Data Preparation

The trainset uses the DIV2K (800). In order to effectively improve the training speed, images are cropped to 480 * 480 images by running script extract_subimages.py, and the dataloader will further randomly crop the images to the GT_size required for training. GT_size defaults to 128/192/256 (×2/×3/×4).

python extract_subimages.py

The input and output paths of cropped pictures can be modify in this script. Default location: ./datasets/DIV2K.

📜Implementation Details

Patches of 64 × 64 pixels are randomly cropped from LR images as input. The model is optimized by minimizing the L1 loss through the Adam optimizer with β1 = 0.9, β2 = 0.999. The initial learning rate is set to be 5×10−4 with a multistep scheduler in 500k and is reduced by half at the (250k,400k,450k,475k)-th iterations.

Hyperparameter	Configuration
BAG	4
BRAB(Each BAG)	2
Feature Channel Dimension	52
Minibatch Size(train)	64

🚀Train

▶️ You can change the training strategy by modifying the configuration file. The default configuration files are included in ./options/train/BGAI. Take one GPU as the example.

### Train ###
### BGAI ###
python train.py -opt ./options/train/BGAI/train_BGAI_x2.yml --auto_resume  # ×2
python train.py -opt ./options/train/BGAI/train_BGAI_x3.yml --auto_resume  # ×3
python train.py -opt ./options/train/BGAI/train_BGAI_x4.yml --auto_resume  # ×4

For more training commands, please check the docs in BasicSR

🚀Test

▶️ You can modify the configuration file about the test, which is located in ./options/test/BGAI. At the same time, you can change the benchmark datasets and modify the path of the pre-train model. The pre-training weights for our model are saved in ./Pretrain_Models.

### Test ###
### BGAI for Lightweight Image Super-Resolution ###
python basicsr/test.py -opt ./options/test/BGAI/test_BGAI_x2.yml  # ×2
python basicsr/test.py -opt ./options/test/BGAI/test_BGAI_x3.yml  # ×3
python basicsr/test.py -opt ./options/test/BGAI/test_BGAI_x4.yml  # ×4

### BGAI for Large Image Super-Resolution ###
### Flicker2K  Test2K  Test4K  Test8K ###
python basicsr/test.py -opt ./options/test/BGAI/test_BGAI_large.yml  # large image

🚩Results

The inference results on benchmark datasets will be available at Google Drive.

📫Contact

If you have any questions, please feel free to contact us wqabby@xupt.edu.cn and bolttt@stu.xupt.edu.cn.