Enhanced_Zero-DCE

Enhanced Zero-DCE is CSE graduation project focused on augmenting the performance and quality of the original Zero-DCE model.

Original Zero-DCE

original project page

Enhancement tasks

Tasks page

Requirements

Install all requirements

pip install -r requirements.txt

Zero-DCE

The basics are same as original Zero-DCE project

1. Python 3.7
2. Pytorch 1.0.0
3. opencv
4. torchvision 0.2.1
5. cuda 10.0

LPIPS

6. lpips

SAM

SAM project page

7. segment-anything
8. jupyter_bbox_widget
9. roboflow
10. dataclasses-json
11. supervision

Train

1. Before you train, download training dataset first. Google drive or baidu cloud [password: 1234]
2. unzip and put the downloaded "train_data" folder to "sample_data" folder.
3. Run the following.

python train.py

Test

python infer.py

Test with SAM (Beta)

python infer.py --with_sam

Default checkpoint the SAM uses is sam_vit_h_4b8939.pth

If you want use other checkpoint, download from SAM project page above and run the following.

python infer.py --with_sam --checkpoint {path to your checkpoint}