基于隐式扩散的重新模糊增强方法python实现源码+项目运行说明.zip

# Training of Stripformer We train Stripformer with ID-Blau in two stages: Relative paths should be defined based on the root directory path. - We pre-train Stripformer for 1000 epochs. - Run the following command. ``` CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node 8 Stripformer/deblur_train_pretrained.py --only_use_generate_data --generate_path ./dataset/GOPRO_Large_Reblur ``` After pretraining, we proceed with finetuning based on the original configuration of Stripformer. - We use the pretrained weights from the 500th epoch for finetuning to avoid excessively long training times. Alternatively, you can choose weights from later epochs if needed. - First use a patch size of 256x256 to train Stripformer for 3000 epochs. - Run the following command. ``` CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node 8 Stripformer/deblur_train.py --resume ./experiments/Stripformer_pretrained/epoch_500_Stripformer_pretrained.pth ``` - After 3000 epochs, we keep training Stripformer for 1000 epochs with patch size 512x512 - Run the following command. ``` CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 python -m torch.distributed.launch --nproc_per_node 8 Stripformer/deblur_train.py --resume ./experiments/Stripformer_first_stage/final_Stripformer_first_stage.pth ```