README.md

Datasets

More details of datasets could be found in section 4.1 of my thesis.

Deblur

• GoPro (train & test):
  • Training set: GoPro train
  • Validation & Testing set: GoPro test
• HIDE(test):
  • Testing set: HIDE test

Dehaze

• NH-haze (train):
  • Training set: NH-haze
• D-haze (test):
  • Validation & Testing set: D-haze

Denoise (gaussian)

• DIV2K (train):
  • Training & Validation set: DIV2K train val
• CBSD68 (test):
  • Testing set: CBSD68
• Kodak24 (test)
  • Testing set: Kodak24
• Set5 (test):
  • Testing set: Set5

Denoise (realworld)

• SIDD small (train & test):
  • Training & Validation set: SIDD train benchmark
  • Testing set: SIDD benchmark
• DND (test):
  • Testing set: DND

Deraindrop

• DeRainDrop (train & test):
  • Training & Validation set: Deraindrop train test
  • Testing set: Deraindrop testA testB

Derainstreak

• Rain13k (train):
  • Training set: Rain13k
• Test2800 (validation & test):
  • Testing set: Test2800
• Rain100H (test):
  • Testing set: Rain100H
• Rain100L (test):
  • Testing set: Rain100L
• Test100 (test):
  • Testing set: Test100

LLEnhancement

• LOL (train & test):
  • Training & Validation set: LOL train eval
  • Testing set: LOL eval

Retouching

• MIT-5K (train & test):
  • Training & Validation set: MIT-5K
  • Testing set: MIT-5K

Preprocess

Crop the patches by generate_patches.py for each dataset according to the following table.
Note that the center crop for validation is build-in in here, you don't need to crop them manually.

Tree

After downloading datasets from above links and finishing the preprocess. Your directory structure should look like this

Training tree...

datasets  # datasets root
  ├── train                    # training data root (have training and validation set)  
        ├── DIV2K              # Denoising training dataset
        |    ├── test          # validation set
        |    |    ├── input    # noise images set
        |    |    └── target   # gt images set 
        |    └── train         # training set
        |         ├── input    # noise images set
        |         └── target   # gt images set 
        |
        ├── FiveK              # Retouching training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |  
        ├── GoPro              # Deblurring training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |  
        ├── LOL                # LLEnhancement training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |    
        ├── NH-Haze            # Dehaze training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |    
        ├── Rain13k            # Derainstreak training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |      
        ├── Raindrop           # Deraindrop training dataset
        |    ├── test
        |    |    ├── input
        |    |    └── target
        |    └── train
        |         ├── input
        |         └── target
        |      
        └── SIDD               # Denoising training dataset
             ├── test
             |    ├── input
             |    └── target
             └── train
                  ├── input
                  └── target