How goutamgmb/deep-burst-sr works
Deep-Burst-SR is a RAW burst super-resolution research toolkit. Its main workflow prepares real or synthetic burst data, trains a model that aligns and fuses frames into a denoised high-resolution RGB image, and evaluates or saves predictions. The graph also includes documented dataset and evaluation branches. Some subsystem connections are documented in the README or SOURCE INDEX rather than visible excerpts; those relationships are cited accordingly.
Main entry points: default_synthetic.py Training settings · burstsr/compute_score.py BurstSR evaluation · synburst/compute_score.py Synthetic evaluation · dbsrnet.py DBSR network
Components 24
Data preparation
- Synthetic processing
data/processing.py - Burst generation
data/synthetic_burst_generation.py - Camera simulation
data/camera_pipeline.py - Data transforms
data/transforms.py - Zurich RAW2RGB
dataset/zurichraw2rgb.py - BurstSR dataset
dataset/burstsr_dataset.py - Synthetic validation
dataset/synthetic_burst_val_set.py
Super-resolution model
- DBSR network
models/dbsr/dbsrnet.py - Optical-flow alignment
models/alignment/pwcnet.py - Correlation layer
external/pwcnet/correlation/correlation.py - Frame fusion
models/dbsr/merging.py - Training losses
models/loss/image_quality_v2.py
Training
- Training entry
run_training.py - Training settings
train_settings/dbsr/default_synthetic.py - Trainer
trainers/base_trainer.py - DBSR actors
actors/dbsr_actors.py
Evaluation
- Evaluation networks
evaluation/common_utils/network_param.py - BurstSR evaluation
evaluation/burstsr/compute_score.py - Synthetic evaluation
evaluation/synburst/compute_score.py - Save predictions
evaluation - Visualize results
evaluation - Quality metrics
models/loss/image_quality_v2.py
Other components
- Researcher
- Environment settings
admin/environment.py
Connections 28
- Researcher → Training entry: starts training
- Training entry → Training settings: loads settings
run_training.py - Training settings → Zurich RAW2RGB: selects dataset
train_settings/dbsr/default_synthetic.py - Training settings → Synthetic processing: configures processing
train_settings/dbsr/default_synthetic.py - Training settings → Data transforms: configures transforms
train_settings/dbsr/default_synthetic.py - Zurich RAW2RGB → Synthetic processing: provides images
- Synthetic processing → Burst generation: generates bursts
data/processing.py - Burst generation → Camera simulation: simulates camera
data/synthetic_burst_generation.py - Training settings → Trainer: configures training
train_settings/dbsr/default_synthetic.py - Trainer → DBSR actors: runs training actor
- DBSR actors → DBSR network: runs network
actors/dbsr_actors.py - DBSR actors → Training losses: computes loss
actors/dbsr_actors.py - DBSR network → Optical-flow alignment: aligns frames
models/dbsr/dbsrnet.py - Optical-flow alignment → Correlation layer: uses correlation
models/alignment/pwcnet.py - DBSR network → Frame fusion: fuses frames
models/dbsr/dbsrnet.py - BurstSR dataset → Environment settings: gets data path
dataset/burstsr_dataset.py - Synthetic validation → Environment settings: gets data path
dataset/synthetic_burst_val_set.py - Researcher → BurstSR evaluation: evaluates BurstSR
- Researcher → Synthetic evaluation: evaluates synthetic set
- BurstSR evaluation → BurstSR dataset: reads dataset
evaluation/burstsr/compute_score.py - Synthetic evaluation → Synthetic validation: reads dataset
evaluation/synburst/compute_score.py - BurstSR evaluation → Optical-flow alignment: aligns predictions
evaluation/burstsr/compute_score.py - BurstSR evaluation → Quality metrics: computes scores
evaluation/burstsr/compute_score.py - Synthetic evaluation → Quality metrics: computes scores
evaluation/synburst/compute_score.py - BurstSR evaluation → Evaluation networks: loads experiment networks
- Synthetic evaluation → Evaluation networks: loads experiment networks
- Researcher → Save predictions: saves predictions
- Researcher → Visualize results: views results