Transformer for Burst Image Super-Resolution (TBSR)
This is the official PyTorch implementation of TBSR.
Our team received 2nd place (real data track) and 3rd place (synthetic track) in NTIRE 2022 Burst Super-Resolution Challenge (CVPRW 2022).
Please see more detailed descriptions of TBSR from TBSR.pdf
.
1. Framework
Overview of the network architecture for TBSR.
2. Preparation
-
Prerequisites
- Python 3.x and PyTorch 1.6.
- OpenCV, NumPy, Pillow, CuPy, tqdm, lpips, scikit-image and tensorboardX.
-
Dataset
- Please see this competition description for the download and use of datasets.
3. Quick Start
-
Training
-
For track 1, modify
dataroot
insh train_track1.sh
and then run: -
For track 2, modify
dataroot
insh train_track2.sh
and then run:
-
-
Testing
-
The pre-trained models can be downloaded. You need to put the two folders in the
./ckpt/
folder. -
For track 1, modify
dataroot
insh test_track1.sh
and then run: -
For track 2, modify
dataroot
insh test_track2.sh
and then run:
-
-
Note
- You can specify which GPU to use by
--gpu_ids
, e.g.,--gpu_ids 0,1
,--gpu_ids 3
,--gpu_ids -1
(for CPU mode). In the default setting, all GPUs are used. - You can refer to options for more arguments.
- You can specify which GPU to use by