Faster-to-See-in-the-Dark

It's a faster method for low-light image processing. This method can improve processing speed by at least 40%.

It also uses See-in-the-Dark (SID) dataset.

Setup

It uses the same dataset as Learning-to-See-in-the-Dark, so you can only download the model to start.

Use the following command to download the dataset or trained model and follow the prompts to select the downloaded content:

python utils/download.py

Also you can download directly from Google Drive:

• Improved Model (Multi): Data and Meta
• Basic Model (Unet): Data and Meta
• Dataset: Sony

Confirm the data structure is as follows:

~/
  dataset/
    Sony/
      long/
        .../
      short/
        .../
  checkpoint/
    Multi/
      model.ckpt.data-00000-of-00001
      model.ckpt.meta
    Unet/
      model.ckpt.data-00000-of-00001
      model.ckpt.meta

Start

Use the following command to train the model:

python train.py

Use the following commands to test:

python test.py

Use the following commands to compare the improved method with basic method:

python compare.py

Result

Use i5-8500 and GTX-1070 to compare: