BSRD (Burst Super-Resolution with Diffsuion Model)

This is the official repository for the following paper:

Kyotaro Tokoro, Kazutoshi Akita, and Norimichi Ukita

Burst Super-Resolution with Diffusion Models for Improving Perceptual Quality, IJCNN2024

SyntheticBurst Dataset

Training

• Download Zurich RAW to RGB dataset and place it in './synth/zurich-raw-to-rgb'.

sh train_synth.sh

Evaluation

• Download syn_burst_val (or DBSR(dataset)) and place it in './synth/SyntheticBurstVal'.
• Download BIPNet Trained model (or BIPNet) and place it in './Trained_models/synth/BIPNet.ckpt'.
• Download Burstormer Trained model (or Burstormer) and place it in './Trained_models/synth/Burstormer.ckpt'.
• Download BSRD Trained model and place it in './Trained_models/synth/BSRD_synth.pt'.

sh sample_synth.sh

BurstSR dataset

Training

• Download BurstSR train and validation set.

sh train_real.sh

Evaluation

• Download burstsr_dataset (or DBSR(dataset)).
• Download PWC Trained model (or pwcnet in Burstormer ) and place it in './pwcnet/pwcnet-network-default.pth'.
• Download BIPNet Trained model (or BIPNet) and place it in './Trained_models/real/BIPNet.ckpt'.
• Download Burstormer Trained model (or Burstormer) and place it in './Trained_models/real/Burstormer.ckpt'.
• Download BSRD Trained model and place it in './Trained_models/real/BSRD_real.pt'.

sh sample_real.sh

References

• guided-diffusion
• Burstormer
• BIPNet