RGB-to-IR-Translation-with-GAN

I designed and implemented a generative adversarial network (GAN) to translate optical (RGB) image to infrared (IR) image.

This project was done within the scope of my graduation project.

Check out the project webpage for details

Here is the project report

Requirements

• Python 3.8
• Pytorch 1.6 or higher
• keras-segmentation if you want to use image segmentations
• Opencv
• tqdm
• scikit-image

Usage

Please use the -h command without further ado. This code designed for FLIR thermal image dataset. If you will train with different dataset, you must modify the utils/dataset.py

Train

The dataset directory must have two folder named "rgb", "ir, and also optionally "segment".

dataset
├───ir
├───rgb
└───segment

python train.py -i $dataset_directory

Note that for resumption of the training, use the following command

python train.py -i $dataset_directory -ce $epoch_number_to_continue

Test

python test.py -ce $epoch_number_of_last_saved_model -i $testset_dir

Resize

If you are using FLIR thermal image dataset, keep in mind that images in the dataset have different sizes, and they are unregistered. To deal with this problem, you can use the following command to make the images registered (approximately). Output will be 640x512

python img_resize.py -rgb $RGB_image_dir -ir $IR_image_dir

Acknowledgement

I would like to thank my advisors Prof. Hakan Ali ÇIRPAN and Dr. Sedat ÖZER who assisted me to research such a pleasant field. Also, I would like to thank them for their guidance and help throughout the project.