phoenixtreesky7 / CFEN-ViT-Dehazing

17 stars 3 forks source link

CFEN-ViT-Dehazing

Complementary Feature Enhanced Network with Vision Transformer for Image Dehazing

This repository contains PyTorch code of our paper: Complementary Feature Enhanced Network with Vision Transformer for Image Dehazing

Test Your Datasets

  1. Download the pretained models: Baidu Yun, Passward:cfen

  2. Unzip them into the /checkpoints/xxx/;

  3. The test images (512x512) should be put in [your test data root]/hazy/;

  4. Run the following commands:

    1). Homogeneous dehazing

    (RESIDE-SOTS Dataset)

    python test.py --dataroot [Your testing data root] --name iid_hlgvit_crs_gd4_cfs_v3_reside --n_feats 24 --hidden_dim_ratio 4 --sb --out_all --which_epoch 32

    (O-HAZE Dataset):

    python test.py --dataroot [Your testing data root] --name iid_hlgvit_crs_gd4_cfs_v3_ohaze --n_feats 24 --hidden_dim_ratio 4 --sb --out_all --which_epoch 20

    2). Non-homogeneous dehazing

    (NH-HAZE):

    python test.py --dataroot [Your testing data root] --name iid_hlgvit_crs_gd4_cfs_v3_nhhaze --n_feats 24 --hidden_dim_ratio 4 --sb --out_all --which_epoch 20

    3). Nighttime dehazing

    python test.py --dataroot [Your testing data root] --name iid_hlgvit_crs_gd4_cfs_v3_nighttime --n_feats 24 --hidden_dim_ratio 2 --sb --out_all

    4). Real_world dehazing

    python test.py --dataroot [Your testing data root] --name iid_hlgvit_crs_gd4_cfs_v3_daytime_realworld --n_feats 24 --hidden_dim_ratio 2 --sb --out_all

Results

Hazy image

Real-world Dehazing 0005_real_B

Dehazing result (Ours)

Real-world Dehazing 0005_fake_A

Hazy image

Real-world Dehazing 0061_real_B

Dehazing result (Ours)

Real-world Dehazing 0061_fake_A

Hazy image

Real-world Dehazing 0085_real_B

Dehazing result (Ours)

Real-world Dehazing 0085_fake_A

Hazy image

Real-world Dehazing 0128_real_B

Dehazing result (Ours)

Real-world Dehazing 0128_fake_A

Citation

If you find this code useful for your research, please cite the paper:

Dong Zhao, Jia Li, Hongyu Li, Long Xu, "Complementary Feature Enhanced Network with Vision Transformer for Image Dehazing", Arxiv