jkhu29 / PBVS2022-TISR

3/75(x4) PBVS 2022 Thermal Image Super-Resolution Challenge (TISR)

PBVS @ CVPR 2022

Thermal Images Super-resolution Challenge

requirements:

1. timm
2. tfrecord

usage

SwinIR + WGAN (PSNR: 33.15, SSIM: 0.9133)

SwinIR + WGAN, use L1Loss & TVLoss & SSIMLoss

python make_dataset.py
# if you want to see the images in tfrecord files
# python -W ignore test_data.py

# for paired x4
sh train.sh

# for paired x2, add --scale 2 to train.sh

# for unpaired x2
sh train_cycle.sh

Our Method(PSNR: 34.2, SSIM: 0.9249)

coming soon...

pretrain models

SwinIR x4: Google Drive or Baidu Pad p: wcl1

put it in "./models/x4", then python test.py

test on ubuntu20.04LTS pytorch-1.7.0 RTX3090