The repository contains the official code for the work "Channel Split Convolutional Neural Network (ChaSNet) for Thermal ImageSuper-Resolution" accepted for PBVS-2021 workshop in-conjuction with CVPR-2021 conference.

- Description

- Result

(* x2 results are taken on LR images obtained by bicubic downscaling on MR data)

- Pre-Trained models

The pre-trained model for track-2 (i.e. scaling of x2) is shared with the repository while the pre-trained model for the track-1 (i.e. scaling of x4) can be downloaded from the link.

- Training the model

To train from scratch, you need to set root directory and dataset directory into options/train/train_vkgenPSNR.json file. Then run the following command to start the training.

python train.py -opt PATH-to-json_file

- Testng the model

To test your pre-trained model, you need to set root directory and dataset directory into options/test/test_VKGen.json file. Then run the following command to start the training.

python test.py -opt PATH-to-json_file

To get the SR images using self-assemble technique, you need to run the following line of code.

python self_assemble_test.py -opt PATH-to-json_file

- Requirement of packages

The list of packages required to run the code is given in chasnet.yml file.

We are thankful to Xinntao for their ESRGAN code using which this work has been implemented. For any problem, you may contact at [email protected].