Seeing in the Dark 💡

Dataset:

Sony Dataset

Seeing in the Dark (SID) dataset consists of 5094 pairs of short-exposure and long-exposure images of shape 4240* 2832. The directory long/ corresponds to long-exposure images which serve as ground truth for the model, and the directory short/ corresponds to the short-exposure images captured at different exposure levels (0.03 to 0.1s).

Run locally:

• Clone the repository:

  git clone https://github.com/Computer-Vision-IIITH-2021/project-wandavision.git

• Navigate to the code folder:

  cd see_in_dark

• Download and extract the dataset folders /long and /short in the Sony_test/ folder.

• Create two new folders: mkdir saved_model/

  mkdir test_result_new/

Training from scratch:

• Change the directory locations on the files as per your directory structure.
• Train the model from scratch: python train_Sony.py
• Test the model: python test_Sony.py

Using the pre-trained model:

• Pretrained model is saved as checkpoint_sony_e4000.pth under Sony_test/saved_model
• To test the pretrained model: python test_Sony.py
• Test images are stored in test_results_Sony/ directory.

Get quantitative results:

• Run all cells of the notebook Quantitative_results.ipynb changing the test directory path to where the test results are stored.
• The notebook generates PSNR and SSIM values between pairs of Ground Truth and output images.

Example Output:

Ground Truth	Original Image	Output from our Model

Original paper:

https://cchen156.github.io/paper/18CVPR_SID.pdf

Results (tested on 255 images):

Drive link