Skip to content

README.md

Latest commit

 

History

History
41 lines (30 loc) · 1.84 KB

README.md

File metadata and controls

41 lines (30 loc) · 1.84 KB

Inverse Tone Mapping

Transform LDR (Low Dynamic Range) images to HDR (High Dynamic Range) images using machine learning which improves image contrast and reconstructs areas lost due to underexposure or overexposure.
The machine learning algorithm is a modified version of FHDR

LDR images are images captured with a typical camera. Since their dynamic range is limited, they are unable to capture all brightness values in a scene without overexposure or underexposure.

Example of an HDR image compared to its LDR counterparts:

HDR Image Overexposure Underexposure

GUI

Example output:

Setup

  • Install miniconda
  • conda env create -f environment.yml
  • conda activate fyp
  • pip install dearpygui==1.2.3
  • python src/scripts/main.py

Dataset

Download the dataset from this link.
This zip file contains training data and testing data.
If you want to do your own data augmentation, you can get the full high definition HDR dataset from here. Then use src/scripts/image_augmentation_script.py and src/scripts/split_dataset.py to generate your own training and testing datasets.

Scripts

Check out the scripts at src/scripts/ for doing training, testing, image augmentation, and running TMO and ITMO.

Test images

You can test with images in test_images/demo or datasets/testing_data_ours/ldr. You can also try with images captured from your own camera.