参考图像彩色化网络修改流程（自用版）

一. SGA-Eliminating Gradient Conflict in Reference-based Line-Art Colorization（2022ECCV）

1. 修改config.yml
修改前

  EPOCH: 40
  BATCH_SIZE: 16
  NUM_WORKER : 4
  TRAIN_DIR : 'anime' # 'anime' or 'afhq_cat' or 'afhq_dog' or afhq_wild
1
2
3
4

修改后

  EPOCH: 400
  BATCH_SIZE: 8
  NUM_WORKER : 0
  TRAIN_DIR : 'nighttime' # 'anime' or 'afhq_cat' or 'afhq_dog' or afhq_wild
1
2
3
4

添加

USE_TENSORBOARD : 'True'
1

2. 修改data_loader.py
添加

 elif config['TRAINING_CONFIG']['TRAIN_DIR'] == 'nighttime':
     self.img_dir = r'F:\RefDataset\KAIST\train\refB'
     self.skt_dir = r'F:\RefDataset\KAIST\train\nightA'
     self.data_list = glob.glob(os.path.join(self.img_dir, '*.jpg'))
1
2
3
4

3. 修改model.py

self.gcn3 = Gconv(in_features=channel, out_features=channel)
self.gcn4 = Gconv(in_features=channel, out_features=channel)
1
2

修改为

self.gcn3 = Gconv(in_ch=channel, out_ch=channel)
self.gcn4 = Gconv(in_ch=channel, out_ch=channel)
1
2

二. SCFT-Reference-Based Sketch Image Colorization using Augmented-Self Reference and Dense Semantic Correspondence（2020CVPR）

和SGA修改一致
1. 修改config.yml
修改前

  EPOCH: 40
  BATCH_SIZE: 16
  NUM_WORKER : 4
  TRAIN_DIR : 'anime' # 'anime' or 'afhq_cat' or 'afhq_dog' or afhq_wild
1
2
3
4

修改后

  EPOCH: 400
  BATCH_SIZE: 8
  NUM_WORKER : 0
  TRAIN_DIR : 'nighttime' # 'anime' or 'afhq_cat' or 'afhq_dog' or afhq_wild
1
2
3
4

添加

USE_TENSORBOARD : 'True'
1

2. 修改data_loader.py
添加

 elif config['TRAINING_CONFIG']['TRAIN_DIR'] == 'nighttime':
     self.img_dir = r'F:\RefDataset\KAIST\train\refB'
     self.skt_dir = r'F:\RefDataset\KAIST\train\nightA'
     self.data_list = glob.glob(os.path.join(self.img_dir, '*.jpg'))
1
2
3
4

三.SPADE-Semantic Image Synthesis with Spatially-Adaptive Normalization（2019CVPR）

下载Synchronized-BatchNorm-PyTorch

https://github.com/vacancy/Synchronized-BatchNorm-PyTorch
1

下载后放到./models/networks/

GitHub提供训练和测试代码

python train.py --name [experiment_name] --dataset_mode custom --label_dir [path_to_labels] -- image_dir [path_to_images] --label_nc [num_labels]
python test.py --name [name_of_experiment] --dataset_mode [dataset_mode] --dataroot [path_to_dataset]
1
2

python train.py --name nighttime --dataset_mode custom --label_dir F:\RefDataset\KAIST\train\nightA -- image_dir F:\RefDataset\KAIST\train\refB --label_nc 182
1

问题1：

IndexError: list index out of range
1

更改

python train.py --name nighttime --dataset_mode custom --label_dir F:\RefDataset\KAIST\train\nightA --image_dir F:\RefDataset\KAIST\train\refB --label_nc 182 --no_instance
1

问题2：

RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR
1

更改

python train.py --name nighttime --dataset_mode custom --dataroot F:\RefDataset\iRay\train --label_dir F:\RefDataset\iRay\train\nightA --image_dir F:\RefDataset\iRay\train\refB --no_instance --contain_dontcare_label --label_nc 512 --no_flip --preprocess_mode none --load_size 256 --crop_size 256 --display_freq 5000 --save_epoch_freq 100 --niter 100--gpu_ids 0
python test.py --name nighttime --dataset_mode custom --dataroot /home/sys120-1/cy/ref_based/iRay/train  --label_dir /home/sys120-1/cy/ref_based/iRay/train/nightA --image_dir /home/sys120-1/cy/ref_based/iRay/train/refB --no_instance --contain_dontcare_label --label_nc 512 --no_flip --preprocess_mode none --load_size 256 --crop_size 256 --gpu_ids 0
1
2