Preparing Benchmarks！

• Preparing Benchmarks！
  • Training Datasets
  • Test Datasets
  • Combine Together

Training Datasets

Here, we provide the 7-task setting required for the RAM&RAM++.
In addition, we also offer the commonly used 3-task setting and 5-task setting in academic research, so you can choose the appropriate setting according to your needs!

3-Task Setting

Dataset	Phase	Source	Task for
OTS_BETA	Train	[HomePage]	Dehaze
Rain-100L	Train & Test	[Download]	Derain
BSD400	Train	[Download]	Denoise
WaterlooED	Train	[Download]	Denoise

5-Task Setting

Dataset	Phase	Source	Task for
OTS_BETA	Train	[HomePage]	Dehaze
Rain-100L	Train & Test	[Download]	Derain
BSD400	Train	[Download]	Denoise
WaterlooED	Train	[Download]	Denoise
LOL-v1	Train & Test	[HomePage]	Low Light Enhancement
GoPro	Train & Test	[Download]	Motion Deblur

7 Task Setting (Ours)

Dataset	Phase	Source	Task for
GoPro	Train & Test	[Download]	Motion Deblur
OTS_ALPHA	Train	[Baidu Cloud(f1zz)]	Dehaze
Rain-13k	Train & Test	[Download]	Derain
LOL-v2	Train & Test	[Real Subset Baidu Cloud(65ay)] / [Synthetic Subset Baidu Cloud(b14u)]	Low Light Enhancement
LSDIR	Train & Test	[HomePage]	Denoise DeJPEG DeBlur

Test Datasets

We also provide both in-distribution and out-of-distributiontest sets for various restoration tasks, facilitating the evaluation of model performance and generalization ability.

In-Distribution

Dataset	Phase	Source	Task for
SOTS	Test	[Download]	3/5/7-task Dehaze
Rain-100L	Test	[Download]	3/5-task Derain
Rain13k-val	Test	[Download]	7-task Derain
CBSD68	Test	[Download]	3/5/7-task Denoise
GoPro	Test	[Download]	5/7-task Motion Deblur
LOL-v1	Test	[HomePage]	5/7-task Low Light Enhancement
LSDIR-val	Test	[HomePage]	7-task DeJPEG DeBlur

Out-of-Distribution

Dataset	Phase	Source	Task for
UIEB	Test	[HomePage]	Underwater Enhancement
Urban100	Test	[Download]	OOD Denoise
CDD11-test	Test	[Download]	OOD 7-task (extreme and mixed)

Combine Together

You need to collect required datasets above and place them under the ./datasets Directory.

Symbolic links is a recommended approach, allowing you to place the datasets anywhere you prefer!

The final directory structure will be arranged as:

datasets	              datasets
  |- BSD400	                |- OTS_ALPHA
    |- 2018.jpg	              |- clear
    |- 2092.jpg	              |- depth
    |- ...	                  |- haze
  |- CBSD68	                |- OTS_BETA
    |- CBSD68	              |- clear
      |- noisy5	              |- depth
      |- noisy10	          |- haze
      |- ...	            |- Rain100L
  |- CDD-11_test	          |- norain-1.png
    |- clear	              |- rain-1.png
    |- haze	                  |- ...
    |- ...	                |- rain13k
  |- gopro	                  |- test
    |- test	                  |- train
    |- train                |- SOTS
  |- LOL	                  |- outdoor
    |- test	                |- UIEB
    |- train	              |- raw-890
  |- LOL-v2	                  |- reference-890
    |- Real_captured        |- urban100
    |- Synthetic	          |- urban100_pepper
  |- LSDIR	                  |- urban100_speckle
    |- 0001000	              |- ...
    |- 0002000	            |- WaterlooED
    |- ...	                  |- WaterlooED
  |- LSDIR-val	                 |- 00001.bmp
    |- 0000001.png	             |- ...
    |- 0000002.png	
    |- ...