EDN-GTM: Encoder-Decoder Network with Guided Transmission Map for Effective Image Dehazing

Introduction

Network Architecture:

Requirements

Main dependencies (or equivalent):

• CUDA 10.0
• CUDNN 7.6
• OpenCV
• Tensorflow 1.14.0
• Keras 2.1.3

For other packages, simply run:

$ pip install -r requirements.txt

Test using Pre-trained Weights

Step 1: Download Pre-trained Weights

• Download pre-trained weights from GoogleDrive or HuggingFace
• Pre-trained weights are available for test on: I-HAZE, O-HAZE, Dense-HAZE, NH-HAZE, SOTS-Outdoor datasets (respective to their filenames)
• Create a folder 'weights' to place downloaded weight files

Step 2: Correct Data Paths in test_on_images.py

• Path to pre-trained weight: weight_path
• Path to output directory: output_dir
• Path to folder containing test images: img_src

Step 3: Run Test Script

$ python test_on_images.py

Train

Step 1: Prepare Dataset

• Each image in a clean-hazy image pair must have the same name
• Make Folder 'A' and Folder 'B' containing hazy and clean images, respectively

Step 2: Correct Data Paths in train.py

• Path to folder containing train data: path/to/data
• Note that path/to/data nevigates to the parent directory of 'A' and 'B' like below:

-- path/to/data /
                |- A (containing hazy images)
                |- B (containing clean images)

Step 3: Run Train Script

$ python train.py

Results

A. Quantitative Results (#Params: number of parameters, MACs: multiply-accumulate operations)

Results on I-HAZE & O-HAZE Datasets:

Results on Dense-HAZE & NH-HAZE Datasets

B. Qualitative Results

Results on I-HAZE & O-HAZE Datasets

Results on Dense-HAZE & NH-HAZE Datasets

Results on SOTS-Outdoor & HSTS Datasets

C. Application to Object Detection

Dehazing in Driving Scenes

Visual dehazing results on synthetic hazy scenes:

Visual dehazing results on realistic hazy scenes:

Object Detection

(Red: ground-truth, Green: detection)

Visual dehazing + detection results on synthetic hazy scenes:

Visual dehazing + detection results on realistic hazy scenes:

Citation

@article{tran2022novel,
  title={A novel encoder-decoder network with guided transmission map for single image dehazing},
  author={Tran, Le-Anh and Moon, Seokyong and Park, Dong-Chul},
  journal={Procedia Computer Science},
  volume={204},
  pages={682--689},
  year={2022},
  publisher={Elsevier}
}

@article{tran2024encoder,
  title={Encoder-decoder networks with guided transmission map for effective image dehazing},
  author={Tran, Le-Anh and Park, Dong-Chul},
  journal={The Visual Computer},
  pages={1--24},
  year={2024},
  publisher={Springer}
}

Have fun!

LA Tran