ClarifyNet: A High-Pass and Low-Pass Filtering Based CNN for Single Image Dehazing

This repository contains the source code of our paper, ClarifyNet (publication inprogress in Journal of Systems Architecture).

Dehazing refers to removing the haze and restoring the details from hazy images. In this paper, we propose ClarifyNet, a novel, end-to-end trainable, convolutional neural network architecture for single image dehazing. We note that a high-pass filter detects sharp edges, texture, and other fine details in the image, whereas a low-pass filter detects color and contrast information. Based on this observation, our key idea is to train ClarifyNet on ground-truth haze-free images, low-pass filtered images, and high-pass filtered images. We comprehensively evaluate ClarifyNet on I-HAZE, O-HAZE, Dense-Haze, and NH-HAZE datasets using PSNR and SSIM metrics and compare them with previous works. On Dense-Haze and NH-HAZE datasets, ClarifyNet achieves the highest PSNR and SSIM. For I-HAZE and O-HAZE, ClarifyNet achieves the highest PSNR and remains within the top three techniques on the SSIM metric. ClarifyNet has a model size of 71MB and an inference latency of 1.5 seconds for dehazing an image of size 3996×4096.

Sample Results

---

Visualization of ClarifyNet results on I-HAZE (I1 to I3), O-HAZE (O1 to O3), Dense-Haze (D1 to D3), and NH-HAZE (N1 to N3) datasets

Project Organization

├── LICENSE
├── README.md          <- The top-level README for developers using this project.
├── data               <- The final, canonical data sets for modeling.
│
├── requirements.txt   <- The requirements file for reproducing the analysis environment.
│
├── src                <- Source code for use in this project.
│   ├── __init__.py    <- Makes src a Python module
│   ├── config.py      <- All configuration params
|   ├── util.py        <- All utilities functions
│   │
│   ├── data           <- Script to generate data in required format
│   │   └── dataloader.py
│   │
│   ├── models         <- Scripts to train models and then use trained models to make
│   │   │                 predictions and test performance.
│   │   ├── train_model.py
|   |   ├── test_model.py
|   |   ├── network.py
|   |   └── loss.py
├── inference          <- Inference directory, where predicted masks are stored.
├── weights            <- Weights directory for saving checkpoints.

---

Get Started

---

Dependencies:

pip install -r requirements.txt

(Optional) Conda Environment Configuration

First, create a conda environment

conda create -n clarifynet # python=3
conda activate clarifynet

Now, add dependencies

Now, you can install the required packages.

pip install -r requirements.txt

Dataset

We have used I-HAZE, O-HAZE, Dense-Haze, and NH-HAZE Dataset. Download the dataset from here, and place it in data/ directory.

Training

change the hyperparameters and configuration parameters according to need in src/config.py.

To train clarifynet, Run following command from /src directory.

python models/train_model.py

All the trained checkpoints for pre-training as well as full model training will be saved in /weights.

Above command will train clarifynet for given number of epochs in src/config.py.

Test performance

To test clarifynet with trained model, Run following command from /src directory.

python models/test_model.py

Citation

Yet to be updated
}

License

---

CC BY-NC-ND 4.0