To cope with the challenges that low-light conditions produce in images, photographers usually use the light provided by the camera flash, thus get- ting better illumination on the scene. Nevertheless, non-uniform illumina- tion, very bright regions, or harsh shadows may arise by using the camera flash, especially in low-light conditions. In contrast, ambient images, where the light is evenly distributed, can be seen as the enhancement version of flash images. Using a fully convolutional network we estimate the ambient version of a given flash image. However, instead of an end-to-end model, we propose to first estimate the ratio-log image and then estimate the ambi- ent image using the ratio-log and flash image. Thus recovering high-quality information, and mitigating visual artifacts. Our proposed model generates suitable natural and uniform illumination on the FAID dataset achieving better performance than state-of-the-art methods with SSIM = 0.662, and PSNR = 15.77. We also analyze the components of our model and how they affect the overall performance. Finally, we introduce a metric to measure the similarity of naturalness of illumination between targets and outputs images.
| Flash image | Synthetic ambient image | Ambient image |
|---|---|---|
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
- Linux
- Python 3.6.8
- PyTorch 1.5.0
- NVIDIA GPU + CUDA CuDNN
- Clone this repo
git clone https://github.com/jozech/ratio-log.git
cd ratio-log
- Download the dataset virtual environment
python download_database.py
If you have problems with the script above, you can download it here, then click on the download button. If you use the external URL, you have to put the 'FAIDres' folder inside a directory called 'datasets'. The dataset contains all images of FAID. All images were reduced to 320x240 resolution for our experiments.
├─ flash-to-ambient/
├─ datasets/
├─ MATLAB_codes/
├─ train_model.py
├─ eval_model.py
├─ eval_sample.py
├─ download_database.py
├─ download_model.py
├─ compute_measures.py
├─ imgs/
├─ models/
└─ tools/
- You can evaluate a model with the validation/test dataset, for 1000 epochs of training.
python download_model.py
python eval_model.py --model=RATIOLOG --eval_data=test --load_epoch=1000
- Train our model, with the default configuration.
python train.py --model=RATIOLOG
- To evaluate one single image
python eval_sample.py --sample_dir=FILENAME
If you want to know more about the hyperparameters, please see tools/config.py.